pmap.c revision 2712:f74a135872bc
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/*
23 * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
24 * Use is subject to license terms.
25 */
26
27#pragma ident	"%Z%%M%	%I%	%E% SMI"
28
29#include <stdio.h>
30#include <stdio_ext.h>
31#include <stdlib.h>
32#include <unistd.h>
33#include <ctype.h>
34#include <fcntl.h>
35#include <string.h>
36#include <dirent.h>
37#include <limits.h>
38#include <link.h>
39#include <libelf.h>
40#include <sys/types.h>
41#include <signal.h>
42#include <sys/stat.h>
43#include <sys/mkdev.h>
44#include <sys/mman.h>
45#include <sys/lgrp_user.h>
46#include <libproc.h>
47#include <libzonecfg.h>
48
49#define	KILOBYTE	1024
50#define	MEGABYTE	(KILOBYTE * KILOBYTE)
51#define	GIGABYTE	(KILOBYTE * KILOBYTE * KILOBYTE)
52
53/*
54 * Round up the value to the nearest kilobyte
55 */
56#define	ROUNDUP_KB(x)	(((x) + (KILOBYTE - 1)) / KILOBYTE)
57
58/*
59 * The alignment should be a power of 2.
60 */
61#define	P2ALIGN(x, align)		((x) & -(align))
62
63#define	INVALID_ADDRESS			(uintptr_t)(-1)
64
65struct totals {
66	ulong_t total_size;
67	ulong_t total_swap;
68	ulong_t total_rss;
69	ulong_t total_anon;
70	ulong_t total_locked;
71};
72
73/*
74 * -L option requires per-page information. The information is presented in an
75 * array of page_descr structures.
76 */
77typedef struct page_descr {
78	uintptr_t	pd_start;	/* start address of a page */
79	size_t		pd_pagesize;	/* page size in bytes */
80	lgrp_id_t	pd_lgrp;	/* lgroup of memory backing the page */
81	int		pd_valid;	/* valid page description if non-zero */
82} page_descr_t;
83
84/*
85 * Per-page information for a memory chunk.
86 * The meminfo(2) system call accepts up to MAX_MEMINFO_CNT pages at once.
87 * When we need to scan larger ranges we divide them in MAX_MEMINFO_CNT sized
88 * chunks. The chunk information is stored in the memory_chunk structure.
89 */
90typedef struct memory_chunk {
91	page_descr_t	page_info[MAX_MEMINFO_CNT];
92	uintptr_t	end_addr;
93	uintptr_t	chunk_start;	/* Starting address */
94	uintptr_t	chunk_end;	/* chunk_end is always <= end_addr */
95	size_t		page_size;
96	int		page_index;	/* Current page */
97	int		page_count;	/* Number of pages */
98} memory_chunk_t;
99
100static volatile int interrupt;
101
102typedef int proc_xmap_f(void *, const prxmap_t *, const char *, int, int);
103
104static	int	xmapping_iter(struct ps_prochandle *, proc_xmap_f *, void *,
105    int);
106static	int	rmapping_iter(struct ps_prochandle *, proc_map_f *, void *);
107
108static	int	look_map(void *, const prmap_t *, const char *);
109static	int	look_smap(void *, const prxmap_t *, const char *, int, int);
110static	int	look_xmap(void *, const prxmap_t *, const char *, int, int);
111static	int	look_xmap_nopgsz(void *, const prxmap_t *, const char *,
112    int, int);
113
114static int gather_map(void *, const prmap_t *, const char *);
115static int gather_xmap(void *, const prxmap_t *, const char *, int, int);
116static int iter_map(proc_map_f *, void *);
117static int iter_xmap(proc_xmap_f *, void *);
118static int parse_addr_range(char *, uintptr_t *, uintptr_t *);
119static void mem_chunk_init(memory_chunk_t *, uintptr_t, size_t);
120
121static	int	perr(char *);
122static	void	printK(long, int);
123static	char	*mflags(uint_t);
124
125static size_t get_contiguous_region(memory_chunk_t *, uintptr_t,
126    uintptr_t, size_t, lgrp_id_t *);
127static void	mem_chunk_get(memory_chunk_t *, uintptr_t);
128static lgrp_id_t addr_to_lgrp(memory_chunk_t *, uintptr_t, size_t *);
129static char	*lgrp2str(lgrp_id_t);
130
131static int	address_in_range(uintptr_t, uintptr_t, size_t);
132static size_t	adjust_addr_range(uintptr_t, uintptr_t, size_t,
133    uintptr_t *, uintptr_t *);
134
135static	int	lflag = 0;
136static	int	Lflag = 0;
137static	int	aflag = 0;
138
139/*
140 * The -A address range is represented as a pair of addresses
141 * <start_addr, end_addr>. Either one of these may be unspecified (set to
142 * INVALID_ADDRESS). If both are unspecified, no address range restrictions are
143 * in place.
144 */
145static  uintptr_t start_addr = INVALID_ADDRESS;
146static	uintptr_t end_addr = INVALID_ADDRESS;
147
148static	int	addr_width, size_width;
149static	char	*command;
150static	char	*procname;
151static	struct ps_prochandle *Pr;
152
153static void intr(int);
154
155typedef struct lwpstack {
156	lwpid_t	lwps_lwpid;
157	stack_t	lwps_stack;
158} lwpstack_t;
159
160typedef struct {
161	prxmap_t	md_xmap;
162	prmap_t		md_map;
163	char		*md_objname;
164	boolean_t	md_last;
165	int		md_doswap;
166} mapdata_t;
167
168static	mapdata_t	*maps;
169static	int		map_count;
170static	int		map_alloc;
171
172static	lwpstack_t *stacks = NULL;
173static	uint_t	nstacks = 0;
174
175#define	MAX_TRIES	5
176
177static int
178getstack(void *data, const lwpstatus_t *lsp)
179{
180	int *np = (int *)data;
181
182	if (Plwp_alt_stack(Pr, lsp->pr_lwpid, &stacks[*np].lwps_stack) == 0) {
183		stacks[*np].lwps_stack.ss_flags |= SS_ONSTACK;
184		stacks[*np].lwps_lwpid = lsp->pr_lwpid;
185		(*np)++;
186	}
187
188	if (Plwp_main_stack(Pr, lsp->pr_lwpid, &stacks[*np].lwps_stack) == 0) {
189		stacks[*np].lwps_lwpid = lsp->pr_lwpid;
190		(*np)++;
191	}
192
193	return (0);
194}
195
196/*
197 * We compare the high memory addresses since stacks are faulted in from
198 * high memory addresses to low memory addresses, and our prmap_t
199 * structures identify only the range of addresses that have been faulted
200 * in so far.
201 */
202static int
203cmpstacks(const void *ap, const void *bp)
204{
205	const lwpstack_t *as = ap;
206	const lwpstack_t *bs = bp;
207	uintptr_t a = (uintptr_t)as->lwps_stack.ss_sp + as->lwps_stack.ss_size;
208	uintptr_t b = (uintptr_t)bs->lwps_stack.ss_sp + bs->lwps_stack.ss_size;
209
210	if (a < b)
211		return (1);
212	if (a > b)
213		return (-1);
214	return (0);
215}
216
217
218int
219main(int argc, char **argv)
220{
221	int rflag = 0, sflag = 0, xflag = 0, Fflag = 0;
222	int errflg = 0, Sflag = 0;
223	int rc = 0;
224	int opt;
225	const char *bar8 = "-------";
226	const char *bar16 = "----------";
227	const char *bar;
228	struct rlimit rlim;
229	struct stat64 statbuf;
230	char buf[128];
231	int mapfd;
232
233	if ((command = strrchr(argv[0], '/')) != NULL)
234		command++;
235	else
236		command = argv[0];
237
238	while ((opt = getopt(argc, argv, "arsxSlLFA:")) != EOF) {
239		switch (opt) {
240		case 'a':		/* include shared mappings in -[xS] */
241			aflag = 1;
242			break;
243		case 'r':		/* show reserved mappings */
244			rflag = 1;
245			break;
246		case 's':		/* show hardware page sizes */
247			sflag = 1;
248			break;
249		case 'S':		/* show swap reservations */
250			Sflag = 1;
251			break;
252		case 'x':		/* show extended mappings */
253			xflag = 1;
254			break;
255		case 'l':		/* show unresolved link map names */
256			lflag = 1;
257			break;
258		case 'L':		/* show lgroup information */
259			Lflag = 1;
260			break;
261		case 'F':		/* force grabbing (no O_EXCL) */
262			Fflag = PGRAB_FORCE;
263			break;
264		case 'A':
265			if (parse_addr_range(optarg, &start_addr, &end_addr)
266			    != 0)
267				errflg++;
268			break;
269		default:
270			errflg = 1;
271			break;
272		}
273	}
274
275	argc -= optind;
276	argv += optind;
277
278	if ((Sflag && (xflag || rflag || sflag)) || (xflag && rflag) ||
279	    (aflag && (!xflag && !Sflag)) ||
280	    (Lflag && (xflag || Sflag))) {
281		errflg = 1;
282	}
283
284	if (errflg || argc <= 0) {
285		(void) fprintf(stderr,
286		    "usage:\t%s [-rslF] [-A start[,end]] { pid | core } ...\n",
287		    command);
288		(void) fprintf(stderr,
289		    "\t\t(report process address maps)\n");
290		(void) fprintf(stderr,
291		    "\t%s -L [-rslF] [-A start[,end]] pid ...\n", command);
292		(void) fprintf(stderr,
293		    "\t\t(report process address maps lgroups mappings)\n");
294		(void) fprintf(stderr,
295		    "\t%s -x [-aslF] [-A start[,end]] pid ...\n", command);
296		(void) fprintf(stderr,
297		    "\t\t(show resident/anon/locked mapping details)\n");
298		(void) fprintf(stderr,
299		    "\t%s -S [-alF] [-A start[,end]] { pid | core } ...\n",
300		    command);
301		(void) fprintf(stderr,
302		    "\t\t(show swap reservations)\n\n");
303		(void) fprintf(stderr,
304		    "\t-a: include shared mappings in -[xS] summary\n");
305		(void) fprintf(stderr,
306		    "\t-r: show reserved address maps\n");
307		(void) fprintf(stderr,
308		    "\t-s: show hardware page sizes\n");
309		(void) fprintf(stderr,
310		    "\t-l: show unresolved dynamic linker map names\n");
311		(void) fprintf(stderr,
312		    "\t-F: force grabbing of the target process\n");
313		(void) fprintf(stderr,
314		    "\t-L: show lgroup mappings\n");
315		(void) fprintf(stderr,
316		    "\t-A start,end: limit output to the specified range\n");
317		return (2);
318	}
319
320	/*
321	 * Make sure we'll have enough file descriptors to handle a target
322	 * that has many many mappings.
323	 */
324	if (getrlimit(RLIMIT_NOFILE, &rlim) == 0) {
325		rlim.rlim_cur = rlim.rlim_max;
326		(void) setrlimit(RLIMIT_NOFILE, &rlim);
327		(void) enable_extended_FILE_stdio(-1, -1);
328	}
329
330	while (argc-- > 0) {
331		char *arg;
332		int gcode;
333		psinfo_t psinfo;
334		int tries = 0;
335		int prg_gflags = PGRAB_RDONLY;
336		int prr_flags = 0;
337
338		if (Lflag) {
339			prg_gflags = PGRAB_RETAIN | Fflag;
340			prr_flags = PRELEASE_RETAIN;
341		}
342
343		if ((Pr = proc_arg_grab(arg = *argv++, PR_ARG_ANY,
344		    prg_gflags, &gcode)) == NULL) {
345			(void) fprintf(stderr, "%s: cannot examine %s: %s\n",
346			    command, arg, Pgrab_error(gcode));
347			rc++;
348			continue;
349		}
350
351		procname = arg;		/* for perr() */
352
353		addr_width = (Pstatus(Pr)->pr_dmodel == PR_MODEL_LP64) ? 16 : 8;
354		size_width = (Pstatus(Pr)->pr_dmodel == PR_MODEL_LP64) ? 11 : 8;
355		bar = addr_width == 8 ? bar8 : bar16;
356		(void) memcpy(&psinfo, Ppsinfo(Pr), sizeof (psinfo_t));
357		proc_unctrl_psinfo(&psinfo);
358
359		if (Pstate(Pr) != PS_DEAD) {
360			(void) snprintf(buf, sizeof (buf),
361			    "/proc/%d/map", (int)psinfo.pr_pid);
362			if ((mapfd = open(buf, O_RDONLY)) < 0) {
363				(void) fprintf(stderr, "%s: cannot "
364				    "examine %s: lost control of "
365				    "process\n", command, arg);
366				rc++;
367				Prelease(Pr, prr_flags);
368				continue;
369			}
370		} else {
371			mapfd = -1;
372		}
373
374again:
375		map_count = 0;
376
377		if (Pstate(Pr) == PS_DEAD) {
378			(void) printf("core '%s' of %d:\t%.70s\n",
379			    arg, (int)psinfo.pr_pid, psinfo.pr_psargs);
380
381			if (rflag || sflag || xflag || Sflag || Lflag) {
382				(void) printf("  -%c option is not compatible "
383				    "with core files\n", xflag ? 'x' :
384				    sflag ? 's' : rflag ? 'r' :
385				    Lflag ? 'L' : 'S');
386				Prelease(Pr, prr_flags);
387				rc++;
388				continue;
389			}
390
391		} else {
392			(void) printf("%d:\t%.70s\n",
393			    (int)psinfo.pr_pid, psinfo.pr_psargs);
394		}
395
396		if (Lflag) {
397			/*
398			 * The implementation of -L option creates an agent LWP
399			 * in the target process address space. The agent LWP
400			 * issues meminfo(2) system calls on behalf of the
401			 * target process. If we are interrupted prematurely,
402			 * the target process remains in the stopped state with
403			 * the agent still attached to it. To prevent such
404			 * situation we catch signals from terminal and
405			 * terminate gracefully.
406			 */
407			if (sigset(SIGHUP, SIG_IGN) == SIG_DFL)
408				(void) sigset(SIGHUP, intr);
409			if (sigset(SIGINT, SIG_IGN) == SIG_DFL)
410				(void) sigset(SIGINT, intr);
411			if (sigset(SIGQUIT, SIG_IGN) == SIG_DFL)
412				(void) sigset(SIGQUIT, intr);
413			(void) sigset(SIGPIPE, intr);
414			(void) sigset(SIGTERM, intr);
415		}
416
417		if (!(Pstatus(Pr)->pr_flags & PR_ISSYS)) {
418			struct totals t;
419
420			/*
421			 * Since we're grabbing the process readonly, we need
422			 * to make sure the address space doesn't change during
423			 * execution.
424			 */
425			if (Pstate(Pr) != PS_DEAD) {
426				if (tries++ == MAX_TRIES) {
427					Prelease(Pr, prr_flags);
428					(void) close(mapfd);
429					(void) fprintf(stderr, "%s: cannot "
430					    "examine %s: address space is "
431					    "changing\n", command, arg);
432					continue;
433				}
434
435				if (fstat64(mapfd, &statbuf) != 0) {
436					Prelease(Pr, prr_flags);
437					(void) close(mapfd);
438					(void) fprintf(stderr, "%s: cannot "
439					    "examine %s: lost control of "
440					    "process\n", command, arg);
441					continue;
442				}
443			}
444
445			nstacks = psinfo.pr_nlwp * 2;
446			stacks = calloc(nstacks, sizeof (stacks[0]));
447			if (stacks != NULL) {
448				int n = 0;
449				(void) Plwp_iter(Pr, getstack, &n);
450				qsort(stacks, nstacks, sizeof (stacks[0]),
451				    cmpstacks);
452			}
453
454			(void) memset(&t, 0, sizeof (t));
455
456			if (Pgetauxval(Pr, AT_BASE) != -1L &&
457			    Prd_agent(Pr) == NULL) {
458				(void) fprintf(stderr, "%s: warning: "
459				    "librtld_db failed to initialize; "
460				    "shared library information will not be "
461				    "available\n", command);
462			}
463
464			/*
465			 * Gather data
466			 */
467			if (xflag)
468				rc += xmapping_iter(Pr, gather_xmap, NULL, 0);
469			else if (Sflag)
470				rc += xmapping_iter(Pr, gather_xmap, NULL, 1);
471			else {
472				if (rflag)
473					rc += rmapping_iter(Pr, gather_map,
474					    NULL);
475				else if (sflag)
476					rc += xmapping_iter(Pr, gather_xmap,
477					    NULL, 0);
478				else
479					rc += Pmapping_iter(Pr, gather_map,
480					    NULL);
481			}
482
483			/*
484			 * Ensure mappings are consistent.
485			 */
486			if (Pstate(Pr) != PS_DEAD) {
487				struct stat64 newbuf;
488
489				if (fstat64(mapfd, &newbuf) != 0 ||
490				    memcmp(&newbuf.st_mtim, &statbuf.st_mtim,
491				    sizeof (newbuf.st_mtim)) != 0) {
492					if (stacks != NULL) {
493						free(stacks);
494						stacks = NULL;
495					}
496					goto again;
497				}
498			}
499
500			/*
501			 * Display data.
502			 */
503			if (xflag) {
504				(void) printf("%*s%*s%*s%*s%*s "
505				    "%sMode   Mapped File\n",
506				    addr_width, "Address",
507				    size_width, "Kbytes",
508				    size_width, "RSS",
509				    size_width, "Anon",
510				    size_width, "Locked",
511				    sflag ? "Pgsz " : "");
512
513				rc += iter_xmap(sflag ?  look_xmap :
514				    look_xmap_nopgsz, &t);
515
516				(void) printf("%s%s %s %s %s %s\n",
517				    addr_width == 8 ? "-" : "------",
518				    bar, bar, bar, bar, bar);
519
520				(void) printf("%stotal Kb", addr_width == 16 ?
521				    "        " : "");
522
523				printK(t.total_size, size_width);
524				printK(t.total_rss, size_width);
525				printK(t.total_anon, size_width);
526				printK(t.total_locked, size_width);
527
528				(void) printf("\n");
529
530			} else if (Sflag) {
531				(void) printf("%*s%*s%*s Mode"
532				    " Mapped File\n",
533				    addr_width, "Address",
534				    size_width, "Kbytes",
535				    size_width, "Swap");
536
537				rc += iter_xmap(look_xmap_nopgsz, &t);
538
539				(void) printf("%s%s %s %s\n",
540				    addr_width == 8 ? "-" : "------",
541				    bar, bar, bar);
542
543				(void) printf("%stotal Kb", addr_width == 16 ?
544				    "        " : "");
545
546				printK(t.total_size, size_width);
547				printK(t.total_swap, size_width);
548
549				(void) printf("\n");
550
551			} else {
552
553				if (rflag) {
554					rc += iter_map(look_map, &t);
555				} else if (sflag) {
556					if (Lflag) {
557						(void) printf("%*s %*s %4s"
558						    " %-6s %s %s\n",
559						    addr_width, "Address",
560						    size_width,
561						    "Bytes", "Pgsz", "Mode ",
562						    "Lgrp", "Mapped File");
563						rc += iter_xmap(look_smap, &t);
564					} else {
565						(void) printf("%*s %*s %4s"
566						    " %-6s %s\n",
567						    addr_width, "Address",
568						    size_width,
569						    "Bytes", "Pgsz", "Mode ",
570						    "Mapped File");
571						rc += iter_xmap(look_smap, &t);
572					}
573				} else {
574					rc += iter_map(look_map, &t);
575				}
576
577				(void) printf(" %stotal  %*luK\n",
578				    addr_width == 16 ?
579				    "        " : "",
580				    size_width, t.total_size);
581			}
582
583			if (stacks != NULL) {
584				free(stacks);
585				stacks = NULL;
586			}
587
588		}
589
590		Prelease(Pr, prr_flags);
591		if (mapfd != -1)
592			(void) close(mapfd);
593	}
594
595	return (rc);
596}
597
598static char *
599make_name(struct ps_prochandle *Pr, uintptr_t addr, const char *mapname,
600	char *buf, size_t bufsz)
601{
602	const pstatus_t		*Psp = Pstatus(Pr);
603	const psinfo_t		*pi = Ppsinfo(Pr);
604	char			fname[100];
605	struct stat		statb;
606	int			len;
607	char			zname[ZONENAME_MAX];
608	char			zpath[PATH_MAX];
609	char			objname[PATH_MAX];
610
611	if (!lflag && strcmp(mapname, "a.out") == 0 &&
612	    Pexecname(Pr, buf, bufsz) != NULL)
613		return (buf);
614
615	if (Pobjname(Pr, addr, objname, sizeof (objname)) != NULL) {
616		(void) strncpy(buf, objname, bufsz);
617
618		if (lflag)
619			return (buf);
620
621		if ((len = resolvepath(buf, buf, bufsz)) > 0) {
622			buf[len] = '\0';
623			return (buf);
624		}
625
626		/*
627		 * If the target is in a non-global zone, attempt to prepend
628		 * the zone path in order to give the global-zone caller the
629		 * real path to the file.
630		 */
631		if (getzonenamebyid(pi->pr_zoneid, zname,
632			sizeof (zname)) != -1 && strcmp(zname, "global") != 0 &&
633		    zone_get_zonepath(zname, zpath, sizeof (zpath)) == Z_OK) {
634			(void) strncat(zpath, "/root",
635			    MAXPATHLEN - strlen(zpath));
636
637			if (bufsz <= strlen(zpath))
638				return (NULL);
639
640			(void) strncpy(buf, zpath, bufsz);
641			(void) strncat(buf, objname, bufsz - strlen(zpath));
642		}
643
644		if ((len = resolvepath(buf, buf, bufsz)) > 0) {
645			buf[len] = '\0';
646			return (buf);
647		}
648	}
649
650	if (Pstate(Pr) != PS_DEAD && *mapname != '\0') {
651		(void) snprintf(fname, sizeof (fname), "/proc/%d/object/%s",
652			(int)Psp->pr_pid, mapname);
653		if (stat(fname, &statb) == 0) {
654			dev_t dev = statb.st_dev;
655			ino_t ino = statb.st_ino;
656			(void) snprintf(buf, bufsz, "dev:%lu,%lu ino:%lu",
657				(ulong_t)major(dev), (ulong_t)minor(dev), ino);
658			return (buf);
659		}
660	}
661
662	return (NULL);
663}
664
665static char *
666anon_name(char *name, const pstatus_t *Psp,
667    uintptr_t vaddr, size_t size, int mflags, int shmid)
668{
669	if (mflags & MA_ISM) {
670		if (shmid == -1)
671			(void) snprintf(name, PATH_MAX, "  [ %s shmid=null ]",
672			    (mflags & MA_NORESERVE) ? "ism" : "dism");
673		else
674			(void) snprintf(name, PATH_MAX, "  [ %s shmid=0x%x ]",
675			    (mflags & MA_NORESERVE) ? "ism" : "dism", shmid);
676	} else if (mflags & MA_SHM) {
677		if (shmid == -1)
678			(void) sprintf(name, "  [ shmid=null ]");
679		else
680			(void) sprintf(name, "  [ shmid=0x%x ]", shmid);
681	} else if (vaddr + size > Psp->pr_stkbase &&
682	    vaddr < Psp->pr_stkbase + Psp->pr_stksize) {
683		(void) strcpy(name, "  [ stack ]");
684	} else if ((mflags & MA_ANON) &&
685	    vaddr + size > Psp->pr_brkbase &&
686	    vaddr < Psp->pr_brkbase + Psp->pr_brksize) {
687		(void) strcpy(name, "  [ heap ]");
688	} else {
689		lwpstack_t key, *stk;
690
691		key.lwps_stack.ss_sp = (void *)vaddr;
692		key.lwps_stack.ss_size = size;
693		if (nstacks > 0 &&
694		    (stk = bsearch(&key, stacks, nstacks, sizeof (stacks[0]),
695		    cmpstacks)) != NULL) {
696			(void) snprintf(name, PATH_MAX, "  [ %s tid=%d ]",
697			    (stk->lwps_stack.ss_flags & SS_ONSTACK) ?
698			    "altstack" : "stack",
699			    stk->lwps_lwpid);
700		} else if (Pstate(Pr) != PS_DEAD) {
701			(void) strcpy(name, "  [ anon ]");
702		} else {
703			return (NULL);
704		}
705	}
706
707	return (name);
708}
709
710static int
711rmapping_iter(struct ps_prochandle *Pr, proc_map_f *func, void *cd)
712{
713	char mapname[PATH_MAX];
714	int mapfd, nmap, i, rc;
715	struct stat st;
716	prmap_t *prmapp, *pmp;
717	ssize_t n;
718
719	(void) snprintf(mapname, sizeof (mapname),
720	    "/proc/%d/rmap", (int)Pstatus(Pr)->pr_pid);
721
722	if ((mapfd = open(mapname, O_RDONLY)) < 0 || fstat(mapfd, &st) != 0) {
723		if (mapfd >= 0)
724			(void) close(mapfd);
725		return (perr(mapname));
726	}
727
728	nmap = st.st_size / sizeof (prmap_t);
729	prmapp = malloc((nmap + 1) * sizeof (prmap_t));
730
731	if ((n = pread(mapfd, prmapp, (nmap + 1) * sizeof (prmap_t), 0L)) < 0) {
732		(void) close(mapfd);
733		free(prmapp);
734		return (perr("read rmap"));
735	}
736
737	(void) close(mapfd);
738	nmap = n / sizeof (prmap_t);
739
740	for (i = 0, pmp = prmapp; i < nmap; i++, pmp++) {
741		if ((rc = func(cd, pmp, NULL)) != 0) {
742			free(prmapp);
743			return (rc);
744		}
745	}
746
747	free(prmapp);
748	return (0);
749}
750
751static int
752xmapping_iter(struct ps_prochandle *Pr, proc_xmap_f *func, void *cd, int doswap)
753{
754	char mapname[PATH_MAX];
755	int mapfd, nmap, i, rc;
756	struct stat st;
757	prxmap_t *prmapp, *pmp;
758	ssize_t n;
759
760	(void) snprintf(mapname, sizeof (mapname),
761	    "/proc/%d/xmap", (int)Pstatus(Pr)->pr_pid);
762
763	if ((mapfd = open(mapname, O_RDONLY)) < 0 || fstat(mapfd, &st) != 0) {
764		if (mapfd >= 0)
765			(void) close(mapfd);
766		return (perr(mapname));
767	}
768
769	nmap = st.st_size / sizeof (prxmap_t);
770	nmap *= 2;
771again:
772	prmapp = malloc((nmap + 1) * sizeof (prxmap_t));
773
774	if ((n = pread(mapfd, prmapp, (nmap + 1) * sizeof (prxmap_t), 0)) < 0) {
775		(void) close(mapfd);
776		free(prmapp);
777		return (perr("read xmap"));
778	}
779
780	if (nmap < n / sizeof (prxmap_t)) {
781		free(prmapp);
782		nmap *= 2;
783		goto again;
784	}
785
786	(void) close(mapfd);
787	nmap = n / sizeof (prxmap_t);
788
789	for (i = 0, pmp = prmapp; i < nmap; i++, pmp++) {
790		if ((rc = func(cd, pmp, NULL, i == nmap - 1, doswap)) != 0) {
791			free(prmapp);
792			return (rc);
793		}
794	}
795
796	/*
797	 * Mark the last element.
798	 */
799	if (map_count > 0)
800		maps[map_count - 1].md_last = B_TRUE;
801
802	free(prmapp);
803	return (0);
804}
805
806/*ARGSUSED*/
807static int
808look_map(void *data, const prmap_t *pmp, const char *object_name)
809{
810	struct totals *t = data;
811	const pstatus_t *Psp = Pstatus(Pr);
812	size_t size;
813	char mname[PATH_MAX];
814	char *lname = NULL;
815	size_t	psz = pmp->pr_pagesize;
816	uintptr_t vaddr = pmp->pr_vaddr;
817	uintptr_t segment_end = vaddr + pmp->pr_size;
818	lgrp_id_t lgrp;
819	memory_chunk_t mchunk;
820
821	/*
822	 * If the mapping is not anon or not part of the heap, make a name
823	 * for it.  We don't want to report the heap as a.out's data.
824	 */
825	if (!(pmp->pr_mflags & MA_ANON) ||
826	    segment_end <= Psp->pr_brkbase ||
827	    pmp->pr_vaddr >= Psp->pr_brkbase + Psp->pr_brksize) {
828		lname = make_name(Pr, pmp->pr_vaddr, pmp->pr_mapname,
829		    mname, sizeof (mname));
830	}
831
832	if (lname == NULL &&
833	    ((pmp->pr_mflags & MA_ANON) || Pstate(Pr) == PS_DEAD)) {
834		lname = anon_name(mname, Psp, pmp->pr_vaddr,
835		    pmp->pr_size, pmp->pr_mflags, pmp->pr_shmid);
836	}
837
838	/*
839	 * Adjust the address range if -A is specified.
840	 */
841	size = adjust_addr_range(pmp->pr_vaddr, segment_end, psz,
842	    &vaddr, &segment_end);
843
844	if (size == 0)
845		return (0);
846
847	if (!Lflag) {
848		/*
849		 * Display the whole mapping
850		 */
851		size = ROUNDUP_KB(size);
852
853		(void) printf(lname ?
854		    "%.*lX %*luK %-6s %s\n" :
855		    "%.*lX %*luK %s\n",
856		    addr_width, vaddr,
857		    size_width - 1, size, mflags(pmp->pr_mflags), lname);
858
859		t->total_size += size;
860		return (0);
861	}
862
863	/*
864	 * We need to display lgroups backing physical memory, so we break the
865	 * segment into individual pages and coalesce pages with the same lgroup
866	 * into one "segment".
867	 */
868
869	/*
870	 * Initialize address descriptions for the mapping.
871	 */
872	mem_chunk_init(&mchunk, segment_end, psz);
873	size = 0;
874
875	/*
876	 * Walk mapping (page by page) and display contiguous ranges of memory
877	 * allocated to same lgroup.
878	 */
879	do {
880		size_t		size_contig;
881
882		/*
883		 * Get contiguous region of memory starting from vaddr allocated
884		 * from the same lgroup.
885		 */
886		size_contig = get_contiguous_region(&mchunk, vaddr,
887		    segment_end, pmp->pr_pagesize, &lgrp);
888
889		(void) printf(lname ? "%.*lX %*luK %-6s%s %s\n" :
890		    "%.*lX %*luK %s %s\n",
891		    addr_width, vaddr,
892		    size_width - 1, size_contig / KILOBYTE,
893		    mflags(pmp->pr_mflags),
894		    lgrp2str(lgrp), lname);
895
896		vaddr += size_contig;
897		size += size_contig;
898	} while (vaddr < segment_end && !interrupt);
899
900	/* Update the total size */
901	t->total_size += ROUNDUP_KB(size);
902	return (0);
903}
904
905static void
906printK(long value, int width)
907{
908	if (value == 0)
909		(void) printf(width == 8 ? "       -" : "          -");
910	else
911		(void) printf(" %*lu", width - 1, value);
912}
913
914static const char *
915pagesize(const prxmap_t *pmp)
916{
917	int pagesize = pmp->pr_hatpagesize;
918	static char buf[32];
919
920	if (pagesize == 0) {
921		return ("-"); /* no underlying HAT mapping */
922	}
923
924	if (pagesize >= KILOBYTE && (pagesize % KILOBYTE) == 0) {
925		if ((pagesize % GIGABYTE) == 0)
926			(void) snprintf(buf, sizeof (buf), "%dG",
927			    pagesize / GIGABYTE);
928		else if ((pagesize % MEGABYTE) == 0)
929			(void) snprintf(buf, sizeof (buf), "%dM",
930			    pagesize / MEGABYTE);
931		else
932			(void) snprintf(buf, sizeof (buf), "%dK",
933			    pagesize / KILOBYTE);
934	} else
935		(void) snprintf(buf, sizeof (buf), "%db", pagesize);
936
937	return (buf);
938}
939
940/*ARGSUSED*/
941static int
942look_smap(void *data,
943	const prxmap_t *pmp,
944	const char *object_name,
945	int last, int doswap)
946{
947	struct totals *t = data;
948	const pstatus_t *Psp = Pstatus(Pr);
949	size_t size;
950	char mname[PATH_MAX];
951	char *lname = NULL;
952	const char *format;
953	size_t	psz = pmp->pr_pagesize;
954	uintptr_t vaddr = pmp->pr_vaddr;
955	uintptr_t segment_end = vaddr + pmp->pr_size;
956	lgrp_id_t lgrp;
957	memory_chunk_t mchunk;
958
959	/*
960	 * If the mapping is not anon or not part of the heap, make a name
961	 * for it.  We don't want to report the heap as a.out's data.
962	 */
963	if (!(pmp->pr_mflags & MA_ANON) ||
964	    pmp->pr_vaddr + pmp->pr_size <= Psp->pr_brkbase ||
965	    pmp->pr_vaddr >= Psp->pr_brkbase + Psp->pr_brksize) {
966		lname = make_name(Pr, pmp->pr_vaddr, pmp->pr_mapname,
967		    mname, sizeof (mname));
968	}
969
970	if (lname == NULL &&
971	    ((pmp->pr_mflags & MA_ANON) || Pstate(Pr) == PS_DEAD)) {
972		lname = anon_name(mname, Psp, pmp->pr_vaddr,
973		    pmp->pr_size, pmp->pr_mflags, pmp->pr_shmid);
974	}
975
976	/*
977	 * Adjust the address range if -A is specified.
978	 */
979	size = adjust_addr_range(pmp->pr_vaddr, segment_end, psz,
980	    &vaddr, &segment_end);
981
982	if (size == 0)
983		return (0);
984
985	if (!Lflag) {
986		/*
987		 * Display the whole mapping
988		 */
989		if (lname != NULL)
990			format = "%.*lX %*luK %4s %-6s %s\n";
991		else
992			format = "%.*lX %*luK %4s %s\n";
993
994		size = ROUNDUP_KB(size);
995
996		(void) printf(format, addr_width, vaddr, size_width - 1, size,
997		    pagesize(pmp), mflags(pmp->pr_mflags), lname);
998
999		t->total_size += size;
1000		return (0);
1001	}
1002
1003	if (lname != NULL)
1004		format = "%.*lX %*luK %4s %-6s%s %s\n";
1005	else
1006		format = "%.*lX %*luK %4s%s %s\n";
1007
1008	/*
1009	 * We need to display lgroups backing physical memory, so we break the
1010	 * segment into individual pages and coalesce pages with the same lgroup
1011	 * into one "segment".
1012	 */
1013
1014	/*
1015	 * Initialize address descriptions for the mapping.
1016	 */
1017	mem_chunk_init(&mchunk, segment_end, psz);
1018	size = 0;
1019
1020	/*
1021	 * Walk mapping (page by page) and display contiguous ranges of memory
1022	 * allocated to same lgroup.
1023	 */
1024	do {
1025		size_t		size_contig;
1026
1027		/*
1028		 * Get contiguous region of memory starting from vaddr allocated
1029		 * from the same lgroup.
1030		 */
1031		size_contig = get_contiguous_region(&mchunk, vaddr,
1032		    segment_end, pmp->pr_pagesize, &lgrp);
1033
1034		(void) printf(format, addr_width, vaddr,
1035		    size_width - 1, size_contig / KILOBYTE,
1036		    pagesize(pmp), mflags(pmp->pr_mflags),
1037		    lgrp2str(lgrp), lname);
1038
1039		vaddr += size_contig;
1040		size += size_contig;
1041	} while (vaddr < segment_end && !interrupt);
1042
1043	t->total_size += ROUNDUP_KB(size);
1044	return (0);
1045}
1046
1047#define	ANON(x)	((aflag || (((x)->pr_mflags & MA_SHARED) == 0)) ? \
1048	    ((x)->pr_anon) : 0)
1049
1050/*ARGSUSED*/
1051static int
1052look_xmap(void *data,
1053	const prxmap_t *pmp,
1054	const char *object_name,
1055	int last, int doswap)
1056{
1057	struct totals *t = data;
1058	const pstatus_t *Psp = Pstatus(Pr);
1059	char mname[PATH_MAX];
1060	char *lname = NULL;
1061	char *ln;
1062
1063	/*
1064	 * If the mapping is not anon or not part of the heap, make a name
1065	 * for it.  We don't want to report the heap as a.out's data.
1066	 */
1067	if (!(pmp->pr_mflags & MA_ANON) ||
1068	    pmp->pr_vaddr + pmp->pr_size <= Psp->pr_brkbase ||
1069	    pmp->pr_vaddr >= Psp->pr_brkbase + Psp->pr_brksize) {
1070		lname = make_name(Pr, pmp->pr_vaddr, pmp->pr_mapname,
1071		    mname, sizeof (mname));
1072	}
1073
1074	if (lname != NULL) {
1075		if ((ln = strrchr(lname, '/')) != NULL)
1076			lname = ln + 1;
1077	} else if ((pmp->pr_mflags & MA_ANON) || Pstate(Pr) == PS_DEAD) {
1078		lname = anon_name(mname, Psp, pmp->pr_vaddr,
1079		    pmp->pr_size, pmp->pr_mflags, pmp->pr_shmid);
1080	}
1081
1082	(void) printf("%.*lX", addr_width, (ulong_t)pmp->pr_vaddr);
1083
1084	printK(ROUNDUP_KB(pmp->pr_size), size_width);
1085	printK(pmp->pr_rss * (pmp->pr_pagesize / KILOBYTE), size_width);
1086	printK(ANON(pmp) * (pmp->pr_pagesize / KILOBYTE), size_width);
1087	printK(pmp->pr_locked * (pmp->pr_pagesize / KILOBYTE), size_width);
1088	(void) printf(lname ? " %4s %-6s %s\n" : " %4s %s\n",
1089	    pagesize(pmp), mflags(pmp->pr_mflags), lname);
1090
1091	t->total_size += ROUNDUP_KB(pmp->pr_size);
1092	t->total_rss += pmp->pr_rss * (pmp->pr_pagesize / KILOBYTE);
1093	t->total_anon += ANON(pmp) * (pmp->pr_pagesize / KILOBYTE);
1094	t->total_locked += (pmp->pr_locked * (pmp->pr_pagesize / KILOBYTE));
1095
1096	return (0);
1097}
1098
1099/*ARGSUSED*/
1100static int
1101look_xmap_nopgsz(void *data,
1102	const prxmap_t *pmp,
1103	const char *object_name,
1104	int last, int doswap)
1105{
1106	struct totals *t = data;
1107	const pstatus_t *Psp = Pstatus(Pr);
1108	char mname[PATH_MAX];
1109	char *lname = NULL;
1110	char *ln;
1111	static uintptr_t prev_vaddr;
1112	static size_t prev_size;
1113	static offset_t prev_offset;
1114	static int prev_mflags;
1115	static char *prev_lname;
1116	static char prev_mname[PATH_MAX];
1117	static ulong_t prev_rss;
1118	static ulong_t prev_anon;
1119	static ulong_t prev_locked;
1120	static ulong_t prev_swap;
1121	int merged = 0;
1122	static int first = 1;
1123	ulong_t swap = 0;
1124	int kperpage;
1125
1126	/*
1127	 * Calculate swap reservations
1128	 */
1129	if (pmp->pr_mflags & MA_SHARED) {
1130		if (aflag && (pmp->pr_mflags & MA_NORESERVE) == 0) {
1131			/* Swap reserved for entire non-ism SHM */
1132			swap = pmp->pr_size / pmp->pr_pagesize;
1133		}
1134	} else if (pmp->pr_mflags & MA_NORESERVE) {
1135		/* Swap reserved on fault for each anon page */
1136		swap = pmp->pr_anon;
1137	} else if (pmp->pr_mflags & MA_WRITE) {
1138		/* Swap reserve for entire writable segment */
1139		swap = pmp->pr_size / pmp->pr_pagesize;
1140	}
1141
1142	/*
1143	 * If the mapping is not anon or not part of the heap, make a name
1144	 * for it.  We don't want to report the heap as a.out's data.
1145	 */
1146	if (!(pmp->pr_mflags & MA_ANON) ||
1147	    pmp->pr_vaddr + pmp->pr_size <= Psp->pr_brkbase ||
1148	    pmp->pr_vaddr >= Psp->pr_brkbase + Psp->pr_brksize) {
1149		lname = make_name(Pr, pmp->pr_vaddr, pmp->pr_mapname,
1150		    mname, sizeof (mname));
1151	}
1152
1153	if (lname != NULL) {
1154		if ((ln = strrchr(lname, '/')) != NULL)
1155			lname = ln + 1;
1156	} else if ((pmp->pr_mflags & MA_ANON) || Pstate(Pr) == PS_DEAD) {
1157		lname = anon_name(mname, Psp, pmp->pr_vaddr,
1158		    pmp->pr_size, pmp->pr_mflags, pmp->pr_shmid);
1159	}
1160
1161	kperpage = pmp->pr_pagesize / KILOBYTE;
1162
1163	t->total_size += ROUNDUP_KB(pmp->pr_size);
1164	t->total_rss += pmp->pr_rss * kperpage;
1165	t->total_anon += ANON(pmp) * kperpage;
1166	t->total_locked += pmp->pr_locked * kperpage;
1167	t->total_swap += swap * kperpage;
1168
1169	if (first == 1) {
1170		first = 0;
1171		prev_vaddr = pmp->pr_vaddr;
1172		prev_size = pmp->pr_size;
1173		prev_offset = pmp->pr_offset;
1174		prev_mflags = pmp->pr_mflags;
1175		if (lname == NULL) {
1176			prev_lname = NULL;
1177		} else {
1178			(void) strcpy(prev_mname, lname);
1179			prev_lname = prev_mname;
1180		}
1181		prev_rss = pmp->pr_rss * kperpage;
1182		prev_anon = ANON(pmp) * kperpage;
1183		prev_locked = pmp->pr_locked * kperpage;
1184		prev_swap = swap * kperpage;
1185		if (last == 0) {
1186			return (0);
1187		}
1188		merged = 1;
1189	} else if (prev_vaddr + prev_size == pmp->pr_vaddr &&
1190	    prev_mflags == pmp->pr_mflags &&
1191	    ((prev_mflags & MA_ISM) ||
1192		prev_offset + prev_size == pmp->pr_offset) &&
1193	    ((lname == NULL && prev_lname == NULL) ||
1194		(lname != NULL && prev_lname != NULL &&
1195		    strcmp(lname, prev_lname) == 0))) {
1196		prev_size += pmp->pr_size;
1197		prev_rss += pmp->pr_rss * kperpage;
1198		prev_anon += ANON(pmp) * kperpage;
1199		prev_locked += pmp->pr_locked * kperpage;
1200		prev_swap += swap * kperpage;
1201		if (last == 0) {
1202			return (0);
1203		}
1204		merged = 1;
1205	}
1206
1207	(void) printf("%.*lX", addr_width, (ulong_t)prev_vaddr);
1208	printK(ROUNDUP_KB(prev_size), size_width);
1209
1210	if (doswap)
1211		printK(prev_swap, size_width);
1212	else {
1213		printK(prev_rss, size_width);
1214		printK(prev_anon, size_width);
1215		printK(prev_locked, size_width);
1216	}
1217	(void) printf(prev_lname ? " %-6s %s\n" : "%s\n",
1218	    mflags(prev_mflags), prev_lname);
1219
1220	if (last == 0) {
1221		prev_vaddr = pmp->pr_vaddr;
1222		prev_size = pmp->pr_size;
1223		prev_offset = pmp->pr_offset;
1224		prev_mflags = pmp->pr_mflags;
1225		if (lname == NULL) {
1226			prev_lname = NULL;
1227		} else {
1228			(void) strcpy(prev_mname, lname);
1229			prev_lname = prev_mname;
1230		}
1231		prev_rss = pmp->pr_rss * kperpage;
1232		prev_anon = ANON(pmp) * kperpage;
1233		prev_locked = pmp->pr_locked * kperpage;
1234		prev_swap = swap * kperpage;
1235	} else if (merged == 0) {
1236		(void) printf("%.*lX", addr_width, (ulong_t)pmp->pr_vaddr);
1237		printK(ROUNDUP_KB(pmp->pr_size), size_width);
1238		if (doswap)
1239			printK(swap * kperpage, size_width);
1240		else {
1241			printK(pmp->pr_rss * kperpage, size_width);
1242			printK(ANON(pmp) * kperpage, size_width);
1243			printK(pmp->pr_locked * kperpage, size_width);
1244		}
1245		(void) printf(lname ? " %-6s %s\n" : " %s\n",
1246		    mflags(pmp->pr_mflags), lname);
1247	}
1248
1249	if (last != 0)
1250		first = 1;
1251
1252	return (0);
1253}
1254
1255static int
1256perr(char *s)
1257{
1258	if (s)
1259		(void) fprintf(stderr, "%s: ", procname);
1260	else
1261		s = procname;
1262	perror(s);
1263	return (1);
1264}
1265
1266static char *
1267mflags(uint_t arg)
1268{
1269	static char code_buf[80];
1270	char *str = code_buf;
1271
1272	/*
1273	 * rwxsR
1274	 *
1275	 * r - segment is readable
1276	 * w - segment is writable
1277	 * x - segment is executable
1278	 * s - segment is shared
1279	 * R - segment is mapped MAP_NORESERVE
1280	 *
1281	 */
1282	(void) sprintf(str, "%c%c%c%c%c%c",
1283	    arg & MA_READ ? 'r' : '-',
1284	    arg & MA_WRITE ? 'w' : '-',
1285	    arg & MA_EXEC ? 'x' : '-',
1286	    arg & MA_SHARED ? 's' : '-',
1287	    arg & MA_NORESERVE ? 'R' : '-',
1288	    arg & MA_RESERVED1 ? '*' : ' ');
1289
1290	return (str);
1291}
1292
1293static mapdata_t *
1294nextmap(void)
1295{
1296	mapdata_t *newmaps;
1297	int next;
1298
1299	if (map_count == map_alloc) {
1300		if (map_alloc == 0)
1301			next = 16;
1302		else
1303			next = map_alloc * 2;
1304
1305		newmaps = realloc(maps, next * sizeof (mapdata_t));
1306		if (newmaps == NULL) {
1307			(void) perr("failed to allocate maps");
1308			exit(1);
1309		}
1310		(void) memset(newmaps + map_alloc, '\0',
1311		    (next - map_alloc) * sizeof (mapdata_t));
1312
1313		map_alloc = next;
1314		maps = newmaps;
1315	}
1316
1317	return (&maps[map_count++]);
1318}
1319
1320/*ARGSUSED*/
1321static int
1322gather_map(void *ignored, const prmap_t *map, const char *objname)
1323{
1324	mapdata_t *data;
1325
1326	/* Skip mappings which are outside the range specified by -A */
1327	if (!address_in_range(map->pr_vaddr,
1328		map->pr_vaddr + map->pr_size, map->pr_pagesize))
1329		return (0);
1330
1331	data = nextmap();
1332	data->md_map = *map;
1333	if (data->md_objname != NULL)
1334		free(data->md_objname);
1335	data->md_objname = objname ? strdup(objname) : NULL;
1336
1337	return (0);
1338}
1339
1340/*ARGSUSED*/
1341static int
1342gather_xmap(void *ignored, const prxmap_t *xmap, const char *objname,
1343    int last, int doswap)
1344{
1345	mapdata_t *data;
1346
1347	/* Skip mappings which are outside the range specified by -A */
1348	if (!address_in_range(xmap->pr_vaddr,
1349		xmap->pr_vaddr + xmap->pr_size, xmap->pr_pagesize))
1350		return (0);
1351
1352	data = nextmap();
1353	data->md_xmap = *xmap;
1354	if (data->md_objname != NULL)
1355		free(data->md_objname);
1356	data->md_objname = objname ? strdup(objname) : NULL;
1357	data->md_last = last;
1358	data->md_doswap = doswap;
1359
1360	return (0);
1361}
1362
1363static int
1364iter_map(proc_map_f *func, void *data)
1365{
1366	int i;
1367	int ret;
1368
1369	for (i = 0; i < map_count; i++) {
1370		if (interrupt)
1371			break;
1372		if ((ret = func(data, &maps[i].md_map,
1373		    maps[i].md_objname)) != 0)
1374			return (ret);
1375	}
1376
1377	return (0);
1378}
1379
1380static int
1381iter_xmap(proc_xmap_f *func, void *data)
1382{
1383	int i;
1384	int ret;
1385
1386	for (i = 0; i < map_count; i++) {
1387		if (interrupt)
1388			break;
1389		if ((ret = func(data, &maps[i].md_xmap, maps[i].md_objname,
1390		    maps[i].md_last, maps[i].md_doswap)) != 0)
1391			return (ret);
1392	}
1393
1394	return (0);
1395}
1396
1397/*
1398 * Convert lgroup ID to string.
1399 * returns dash when lgroup ID is invalid.
1400 */
1401static char *
1402lgrp2str(lgrp_id_t lgrp)
1403{
1404	static char lgrp_buf[20];
1405	char *str = lgrp_buf;
1406
1407	(void) sprintf(str, lgrp == LGRP_NONE ? "   -" : "%4d", lgrp);
1408	return (str);
1409}
1410
1411/*
1412 * Parse address range specification for -A option.
1413 * The address range may have the following forms:
1414 *
1415 * address
1416 *	start and end is set to address
1417 * address,
1418 *	start is set to address, end is set to INVALID_ADDRESS
1419 * ,address
1420 *	start is set to 0, end is set to address
1421 * address1,address2
1422 *	start is set to address1, end is set to address2
1423 *
1424 */
1425static int
1426parse_addr_range(char *input_str, uintptr_t *start, uintptr_t *end)
1427{
1428	char *startp = input_str;
1429	char *endp = strchr(input_str, ',');
1430	ulong_t	s = (ulong_t)INVALID_ADDRESS;
1431	ulong_t e = (ulong_t)INVALID_ADDRESS;
1432
1433	if (endp != NULL) {
1434		/*
1435		 * Comma is present. If there is nothing after comma, the end
1436		 * remains set at INVALID_ADDRESS. Otherwise it is set to the
1437		 * value after comma.
1438		 */
1439		*endp = '\0';
1440		endp++;
1441
1442		if ((*endp != '\0') && sscanf(endp, "%lx", &e) != 1)
1443			return (1);
1444	}
1445
1446	if (startp != NULL) {
1447		/*
1448		 * Read the start address, if it is specified. If the address is
1449		 * missing, start will be set to INVALID_ADDRESS.
1450		 */
1451		if ((*startp != '\0') && sscanf(startp, "%lx", &s) != 1)
1452			return (1);
1453	}
1454
1455	/* If there is no comma, end becomes equal to start */
1456	if (endp == NULL)
1457		e = s;
1458
1459	/*
1460	 * ,end implies 0..end range
1461	 */
1462	if (e != INVALID_ADDRESS && s == INVALID_ADDRESS)
1463		s = 0;
1464
1465	*start = (uintptr_t)s;
1466	*end = (uintptr_t)e;
1467
1468	/* Return error if neither start nor end address were specified */
1469	return (! (s != INVALID_ADDRESS || e != INVALID_ADDRESS));
1470}
1471
1472/*
1473 * Check whether any portion of [start, end] segment is within the
1474 * [start_addr, end_addr] range.
1475 *
1476 * Return values:
1477 *   0 - address is outside the range
1478 *   1 - address is within the range
1479 */
1480static int
1481address_in_range(uintptr_t start, uintptr_t end, size_t psz)
1482{
1483	int rc = 1;
1484
1485	/*
1486	 *  Nothing to do if there is no address range specified with -A
1487	 */
1488	if (start_addr != INVALID_ADDRESS || end_addr != INVALID_ADDRESS) {
1489		/* The segment end is below the range start */
1490		if ((start_addr != INVALID_ADDRESS) &&
1491		    (end < P2ALIGN(start_addr, psz)))
1492			rc = 0;
1493
1494		/* The segment start is above the range end */
1495		if ((end_addr != INVALID_ADDRESS) &&
1496		    (start > P2ALIGN(end_addr + psz, psz)))
1497			rc = 0;
1498	}
1499	return (rc);
1500}
1501
1502/*
1503 * Returns an intersection of the [start, end] interval and the range specified
1504 * by -A flag [start_addr, end_addr]. Unspecified parts of the address range
1505 * have value INVALID_ADDRESS.
1506 *
1507 * The start_addr address is rounded down to the beginning of page and end_addr
1508 * is rounded up to the end of page.
1509 *
1510 * Returns the size of the resulting interval or zero if the interval is empty
1511 * or invalid.
1512 */
1513static size_t
1514adjust_addr_range(uintptr_t start, uintptr_t end, size_t psz,
1515    uintptr_t *new_start, uintptr_t *new_end)
1516{
1517	uintptr_t from;		/* start_addr rounded down */
1518	uintptr_t to;		/* end_addr rounded up */
1519
1520	/*
1521	 * Round down the lower address of the range to the beginning of page.
1522	 */
1523	if (start_addr == INVALID_ADDRESS) {
1524		/*
1525		 * No start_addr specified by -A, the lower part of the interval
1526		 * does not change.
1527		 */
1528		*new_start = start;
1529	} else {
1530		from = P2ALIGN(start_addr, psz);
1531		/*
1532		 * If end address is outside the range, return an empty
1533		 * interval
1534		 */
1535		if (end <  from) {
1536			*new_start = *new_end = 0;
1537			return (0);
1538		}
1539		/*
1540		 * The adjusted start address is the maximum of requested start
1541		 * and the aligned start_addr of the -A range.
1542		 */
1543		*new_start = start < from ? from : start;
1544	}
1545
1546	/*
1547	 * Round up the higher address of the range to the end of page.
1548	 */
1549	if (end_addr == INVALID_ADDRESS) {
1550		/*
1551		 * No end_addr specified by -A, the upper part of the interval
1552		 * does not change.
1553		 */
1554		*new_end = end;
1555	} else {
1556		/*
1557		 * If only one address is specified and it is the beginning of a
1558		 * segment, get information about the whole segment. This
1559		 * function is called once per segment and the 'end' argument is
1560		 * always the end of a segment, so just use the 'end' value.
1561		 */
1562		to = (end_addr == start_addr && start == start_addr) ?
1563		    end :
1564		    P2ALIGN(end_addr + psz, psz);
1565		/*
1566		 * If start address is outside the range, return an empty
1567		 * interval
1568		 */
1569		if (start > to) {
1570			*new_start = *new_end = 0;
1571			return (0);
1572		}
1573		/*
1574		 * The adjusted end address is the minimum of requested end
1575		 * and the aligned end_addr of the -A range.
1576		 */
1577		*new_end = end > to ? to : end;
1578	}
1579
1580	/*
1581	 * Make sure that the resulting interval is legal.
1582	 */
1583	if (*new_end < *new_start)
1584			*new_start = *new_end = 0;
1585
1586	/* Return the size of the interval */
1587	return (*new_end - *new_start);
1588}
1589
1590/*
1591 * Initialize memory_info data structure with information about a new segment.
1592 */
1593static void
1594mem_chunk_init(memory_chunk_t *chunk, uintptr_t end, size_t psz)
1595{
1596	chunk->end_addr = end;
1597	chunk->page_size = psz;
1598	chunk->page_index = 0;
1599	chunk->chunk_start = chunk->chunk_end = 0;
1600}
1601
1602/*
1603 * Create a new chunk of addresses starting from vaddr.
1604 * Pass the whole chunk to pr_meminfo to collect lgroup and page size
1605 * information for each page in the chunk.
1606 */
1607static void
1608mem_chunk_get(memory_chunk_t *chunk, uintptr_t vaddr)
1609{
1610	page_descr_t	*pdp = chunk->page_info;
1611	size_t		psz = chunk->page_size;
1612	uintptr_t	addr = vaddr;
1613	uint64_t	inaddr[MAX_MEMINFO_CNT];
1614	uint64_t	outdata[2 * MAX_MEMINFO_CNT];
1615	uint_t		info[2] = { MEMINFO_VLGRP, MEMINFO_VPAGESIZE };
1616	uint_t		validity[MAX_MEMINFO_CNT];
1617	uint64_t	*dataptr = inaddr;
1618	uint64_t	*outptr = outdata;
1619	uint_t		*valptr = validity;
1620	int 		i, j, rc;
1621
1622	chunk->chunk_start = vaddr;
1623	chunk->page_index = 0;	/* reset index for the new chunk */
1624
1625	/*
1626	 * Fill in MAX_MEMINFO_CNT wotrh of pages starting from vaddr. Also,
1627	 * copy starting address of each page to inaddr array for pr_meminfo.
1628	 */
1629	for (i = 0, pdp = chunk->page_info;
1630	    (i < MAX_MEMINFO_CNT) && (addr <= chunk->end_addr);
1631	    i++, pdp++, dataptr++, addr += psz) {
1632		*dataptr = (uint64_t)addr;
1633		pdp->pd_start = addr;
1634		pdp->pd_lgrp = LGRP_NONE;
1635		pdp->pd_valid = 0;
1636		pdp->pd_pagesize = 0;
1637	}
1638
1639	/* Mark the number of entries in the chunk and the last address */
1640	chunk->page_count = i;
1641	chunk->chunk_end = addr - psz;
1642
1643	if (interrupt)
1644		return;
1645
1646	/* Call meminfo for all collected addresses */
1647	rc = pr_meminfo(Pr, inaddr, i, info, 2, outdata, validity);
1648	if (rc < 0) {
1649		(void) perr("can not get memory information");
1650		return;
1651	}
1652
1653	/* Verify validity of each result and fill in the addrs array */
1654	pdp = chunk->page_info;
1655	for (j = 0; j < i; j++, pdp++, valptr++, outptr += 2) {
1656		/* Skip invalid address pointers */
1657		if ((*valptr & 1) == 0) {
1658			continue;
1659		}
1660
1661		/* Is lgroup information available? */
1662		if ((*valptr & 2) != 0) {
1663			pdp->pd_lgrp = (lgrp_id_t)*outptr;
1664			pdp->pd_valid = 1;
1665		}
1666
1667		/* Is page size informaion available? */
1668		if ((*valptr & 4) != 0) {
1669			pdp->pd_pagesize = *(outptr + 1);
1670		}
1671	}
1672}
1673
1674/*
1675 * Starting from address 'vaddr' find the region with pages allocated from the
1676 * same lgroup.
1677 *
1678 * Arguments:
1679 *	mchunk		Initialized memory chunk structure
1680 *	vaddr		Starting address of the region
1681 *	maxaddr		Upper bound of the region
1682 *	pagesize	Default page size to use
1683 *	ret_lgrp	On exit contains the lgroup ID of all pages in the
1684 *			region.
1685 *
1686 * Returns:
1687 *	Size of the contiguous region in bytes
1688 *	The lgroup ID of all pages in the region in ret_lgrp argument.
1689 */
1690static size_t
1691get_contiguous_region(memory_chunk_t *mchunk, uintptr_t vaddr,
1692    uintptr_t maxaddr, size_t pagesize, lgrp_id_t *ret_lgrp)
1693{
1694	size_t		size_contig = 0;
1695	lgrp_id_t	lgrp;		/* Lgroup of the region start */
1696	lgrp_id_t	curr_lgrp;	/* Lgroup of the current page */
1697	size_t		psz = pagesize;	/* Pagesize to use */
1698
1699	/* Set both lgroup IDs to the lgroup of the first page */
1700	curr_lgrp = lgrp = addr_to_lgrp(mchunk, vaddr, &psz);
1701
1702	/*
1703	 * Starting from vaddr, walk page by page until either the end
1704	 * of the segment is reached or a page is allocated from a different
1705	 * lgroup. Also stop if interrupted from keyboard.
1706	 */
1707	while ((vaddr < maxaddr) && (curr_lgrp == lgrp) && !interrupt) {
1708		/*
1709		 * Get lgroup ID and the page size of the current page.
1710		 */
1711		curr_lgrp = addr_to_lgrp(mchunk, vaddr, &psz);
1712		/* If there is no page size information, use the default */
1713		if (psz == 0)
1714			psz = pagesize;
1715
1716		if (curr_lgrp == lgrp) {
1717			/*
1718			 * This page belongs to the contiguous region.
1719			 * Increase the region size and advance to the new page.
1720			 */
1721			size_contig += psz;
1722			vaddr += psz;
1723		}
1724	}
1725
1726	/* Return the region lgroup ID and the size */
1727	*ret_lgrp = lgrp;
1728	return (size_contig);
1729}
1730
1731/*
1732 * Given a virtual address, return its lgroup and page size. If there is meminfo
1733 * information for an address, use it, otherwise shift the chunk window to the
1734 * vaddr and create a new chunk with known meminfo information.
1735 */
1736static lgrp_id_t
1737addr_to_lgrp(memory_chunk_t *chunk, uintptr_t vaddr, size_t *psz)
1738{
1739	page_descr_t *pdp;
1740	lgrp_id_t lgrp = LGRP_NONE;
1741	int i;
1742
1743	*psz = chunk->page_size;
1744
1745	if (interrupt)
1746		return (0);
1747
1748	/*
1749	 * Is there information about this address? If not, create a new chunk
1750	 * starting from vaddr and apply pr_meminfo() to the whole chunk.
1751	 */
1752	if (vaddr < chunk->chunk_start || vaddr > chunk->chunk_end) {
1753		/*
1754		 * This address is outside the chunk, get the new chunk and
1755		 * collect meminfo information for it.
1756		 */
1757		mem_chunk_get(chunk, vaddr);
1758	}
1759
1760	/*
1761	 * Find information about the address.
1762	 */
1763	pdp = &chunk->page_info[chunk->page_index];
1764	for (i = chunk->page_index; i < chunk->page_count; i++, pdp++) {
1765		if (pdp->pd_start == vaddr) {
1766			if (pdp->pd_valid) {
1767				lgrp = pdp->pd_lgrp;
1768				/*
1769				 * Override page size information if it is
1770				 * present.
1771				 */
1772				if (pdp->pd_pagesize > 0)
1773					*psz = pdp->pd_pagesize;
1774			}
1775			break;
1776		}
1777	}
1778	/*
1779	 * Remember where we ended - the next search will start here.
1780	 * We can query for the lgrp for the same address again, so do not
1781	 * advance index past the current value.
1782	 */
1783	chunk->page_index = i;
1784
1785	return (lgrp);
1786}
1787
1788/* ARGSUSED */
1789static void
1790intr(int sig)
1791{
1792	interrupt = 1;
1793}
1794