xref: /illumos-gate/usr/src/cmd/ptools/pmap/pmap.c (revision 186f7fbf)
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 2008 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #include <stdio.h>
28 #include <stdio_ext.h>
29 #include <stdlib.h>
30 #include <unistd.h>
31 #include <ctype.h>
32 #include <fcntl.h>
33 #include <string.h>
34 #include <dirent.h>
35 #include <limits.h>
36 #include <link.h>
37 #include <libelf.h>
38 #include <sys/types.h>
39 #include <signal.h>
40 #include <sys/stat.h>
41 #include <sys/mkdev.h>
42 #include <sys/mman.h>
43 #include <sys/lgrp_user.h>
44 #include <libproc.h>
45 
46 #include "pmap_common.h"
47 
48 #define	KILOBYTE	1024
49 #define	MEGABYTE	(KILOBYTE * KILOBYTE)
50 #define	GIGABYTE	(KILOBYTE * KILOBYTE * KILOBYTE)
51 
52 /*
53  * Round up the value to the nearest kilobyte
54  */
55 #define	ROUNDUP_KB(x)	(((x) + (KILOBYTE - 1)) / KILOBYTE)
56 
57 /*
58  * The alignment should be a power of 2.
59  */
60 #define	P2ALIGN(x, align)		((x) & -(align))
61 
62 #define	INVALID_ADDRESS			(uintptr_t)(-1)
63 
64 struct totals {
65 	ulong_t total_size;
66 	ulong_t total_swap;
67 	ulong_t total_rss;
68 	ulong_t total_anon;
69 	ulong_t total_locked;
70 };
71 
72 /*
73  * -L option requires per-page information. The information is presented in an
74  * array of page_descr structures.
75  */
76 typedef struct page_descr {
77 	uintptr_t	pd_start;	/* start address of a page */
78 	size_t		pd_pagesize;	/* page size in bytes */
79 	lgrp_id_t	pd_lgrp;	/* lgroup of memory backing the page */
80 	int		pd_valid;	/* valid page description if non-zero */
81 } page_descr_t;
82 
83 /*
84  * Per-page information for a memory chunk.
85  * The meminfo(2) system call accepts up to MAX_MEMINFO_CNT pages at once.
86  * When we need to scan larger ranges we divide them in MAX_MEMINFO_CNT sized
87  * chunks. The chunk information is stored in the memory_chunk structure.
88  */
89 typedef struct memory_chunk {
90 	page_descr_t	page_info[MAX_MEMINFO_CNT];
91 	uintptr_t	end_addr;
92 	uintptr_t	chunk_start;	/* Starting address */
93 	uintptr_t	chunk_end;	/* chunk_end is always <= end_addr */
94 	size_t		page_size;
95 	int		page_index;	/* Current page */
96 	int		page_count;	/* Number of pages */
97 } memory_chunk_t;
98 
99 static volatile int interrupt;
100 
101 typedef int proc_xmap_f(void *, const prxmap_t *, const char *, int, int);
102 
103 static	int	xmapping_iter(struct ps_prochandle *, proc_xmap_f *, void *,
104     int);
105 static	int	rmapping_iter(struct ps_prochandle *, proc_map_f *, void *);
106 
107 static	int	look_map(void *, const prmap_t *, const char *);
108 static	int	look_smap(void *, const prxmap_t *, const char *, int, int);
109 static	int	look_xmap(void *, const prxmap_t *, const char *, int, int);
110 static	int	look_xmap_nopgsz(void *, const prxmap_t *, const char *,
111     int, int);
112 
113 static int gather_map(void *, const prmap_t *, const char *);
114 static int gather_xmap(void *, const prxmap_t *, const char *, int, int);
115 static int iter_map(proc_map_f *, void *);
116 static int iter_xmap(proc_xmap_f *, void *);
117 static int parse_addr_range(char *, uintptr_t *, uintptr_t *);
118 static void mem_chunk_init(memory_chunk_t *, uintptr_t, size_t);
119 
120 static	int	perr(char *);
121 static	void	printK(long, int);
122 static	char	*mflags(uint_t);
123 
124 static size_t get_contiguous_region(memory_chunk_t *, uintptr_t,
125     uintptr_t, size_t, lgrp_id_t *);
126 static void	mem_chunk_get(memory_chunk_t *, uintptr_t);
127 static lgrp_id_t addr_to_lgrp(memory_chunk_t *, uintptr_t, size_t *);
128 static char	*lgrp2str(lgrp_id_t);
129 
130 static int	address_in_range(uintptr_t, uintptr_t, size_t);
131 static size_t	adjust_addr_range(uintptr_t, uintptr_t, size_t,
132     uintptr_t *, uintptr_t *);
133 
134 static	int	lflag = 0;
135 static	int	Lflag = 0;
136 static	int	aflag = 0;
137 
138 /*
139  * The -A address range is represented as a pair of addresses
140  * <start_addr, end_addr>. Either one of these may be unspecified (set to
141  * INVALID_ADDRESS). If both are unspecified, no address range restrictions are
142  * in place.
143  */
144 static  uintptr_t start_addr = INVALID_ADDRESS;
145 static	uintptr_t end_addr = INVALID_ADDRESS;
146 
147 static	int	addr_width, size_width;
148 static	char	*command;
149 static	char	*procname;
150 static	struct ps_prochandle *Pr;
151 
152 static void intr(int);
153 
154 typedef struct {
155 	prxmap_t	md_xmap;
156 	prmap_t		md_map;
157 	char		*md_objname;
158 	boolean_t	md_last;
159 	int		md_doswap;
160 } mapdata_t;
161 
162 static	mapdata_t	*maps;
163 static	int		map_count;
164 static	int		map_alloc;
165 
166 static	lwpstack_t *stacks = NULL;
167 static	uint_t	nstacks = 0;
168 
169 #define	MAX_TRIES	5
170 
171 static int
getstack(void * data,const lwpstatus_t * lsp)172 getstack(void *data, const lwpstatus_t *lsp)
173 {
174 	int *np = (int *)data;
175 
176 	if (Plwp_alt_stack(Pr, lsp->pr_lwpid, &stacks[*np].lwps_stack) == 0) {
177 		stacks[*np].lwps_stack.ss_flags |= SS_ONSTACK;
178 		stacks[*np].lwps_lwpid = lsp->pr_lwpid;
179 		(*np)++;
180 	}
181 
182 	if (Plwp_main_stack(Pr, lsp->pr_lwpid, &stacks[*np].lwps_stack) == 0) {
183 		stacks[*np].lwps_lwpid = lsp->pr_lwpid;
184 		(*np)++;
185 	}
186 
187 	return (0);
188 }
189 
190 int
main(int argc,char ** argv)191 main(int argc, char **argv)
192 {
193 	int rflag = 0, sflag = 0, xflag = 0, Fflag = 0;
194 	int errflg = 0, Sflag = 0;
195 	int rc = 0;
196 	int opt;
197 	const char *bar8 = "-------";
198 	const char *bar16 = "----------";
199 	const char *bar;
200 	struct rlimit rlim;
201 	struct stat64 statbuf;
202 	char buf[128];
203 	int mapfd;
204 	int prg_gflags = PGRAB_RDONLY;
205 	int prr_flags = 0;
206 	boolean_t use_agent_lwp = B_FALSE;
207 
208 	if ((command = strrchr(argv[0], '/')) != NULL)
209 		command++;
210 	else
211 		command = argv[0];
212 
213 	while ((opt = getopt(argc, argv, "arsxSlLFA:")) != EOF) {
214 		switch (opt) {
215 		case 'a':		/* include shared mappings in -[xS] */
216 			aflag = 1;
217 			break;
218 		case 'r':		/* show reserved mappings */
219 			rflag = 1;
220 			break;
221 		case 's':		/* show hardware page sizes */
222 			sflag = 1;
223 			break;
224 		case 'S':		/* show swap reservations */
225 			Sflag = 1;
226 			break;
227 		case 'x':		/* show extended mappings */
228 			xflag = 1;
229 			break;
230 		case 'l':		/* show unresolved link map names */
231 			lflag = 1;
232 			break;
233 		case 'L':		/* show lgroup information */
234 			Lflag = 1;
235 			use_agent_lwp = B_TRUE;
236 			break;
237 		case 'F':		/* force grabbing (no O_EXCL) */
238 			Fflag = PGRAB_FORCE;
239 			break;
240 		case 'A':
241 			if (parse_addr_range(optarg, &start_addr, &end_addr)
242 			    != 0)
243 				errflg++;
244 			break;
245 		default:
246 			errflg = 1;
247 			break;
248 		}
249 	}
250 
251 	argc -= optind;
252 	argv += optind;
253 
254 	if ((Sflag && (xflag || rflag || sflag)) || (xflag && rflag) ||
255 	    (aflag && (!xflag && !Sflag)) ||
256 	    (Lflag && (xflag || Sflag))) {
257 		errflg = 1;
258 	}
259 
260 	if (errflg || argc <= 0) {
261 		(void) fprintf(stderr,
262 		    "usage:\t%s [-rslF] [-A start[,end]] { pid | core } ...\n",
263 		    command);
264 		(void) fprintf(stderr,
265 		    "\t\t(report process address maps)\n");
266 		(void) fprintf(stderr,
267 		    "\t%s -L [-rslF] [-A start[,end]] pid ...\n", command);
268 		(void) fprintf(stderr,
269 		    "\t\t(report process address maps lgroups mappings)\n");
270 		(void) fprintf(stderr,
271 		    "\t%s -x [-aslF] [-A start[,end]] pid ...\n", command);
272 		(void) fprintf(stderr,
273 		    "\t\t(show resident/anon/locked mapping details)\n");
274 		(void) fprintf(stderr,
275 		    "\t%s -S [-alF] [-A start[,end]] { pid | core } ...\n",
276 		    command);
277 		(void) fprintf(stderr,
278 		    "\t\t(show swap reservations)\n\n");
279 		(void) fprintf(stderr,
280 		    "\t-a: include shared mappings in -[xS] summary\n");
281 		(void) fprintf(stderr,
282 		    "\t-r: show reserved address maps\n");
283 		(void) fprintf(stderr,
284 		    "\t-s: show hardware page sizes\n");
285 		(void) fprintf(stderr,
286 		    "\t-l: show unresolved dynamic linker map names\n");
287 		(void) fprintf(stderr,
288 		    "\t-F: force grabbing of the target process\n");
289 		(void) fprintf(stderr,
290 		    "\t-L: show lgroup mappings\n");
291 		(void) fprintf(stderr,
292 		    "\t-A start,end: limit output to the specified range\n");
293 		return (2);
294 	}
295 
296 	/*
297 	 * Make sure we'll have enough file descriptors to handle a target
298 	 * that has many many mappings.
299 	 */
300 	if (getrlimit(RLIMIT_NOFILE, &rlim) == 0) {
301 		rlim.rlim_cur = rlim.rlim_max;
302 		(void) setrlimit(RLIMIT_NOFILE, &rlim);
303 		(void) enable_extended_FILE_stdio(-1, -1);
304 	}
305 
306 	/*
307 	 * The implementation of -L option creates an agent LWP in the target
308 	 * process address space. The agent LWP issues meminfo(2) system calls
309 	 * on behalf of the target process. If we are interrupted prematurely,
310 	 * the target process remains in the stopped state with the agent still
311 	 * attached to it. To prevent such situation we catch signals from
312 	 * terminal and terminate gracefully.
313 	 */
314 	if (use_agent_lwp) {
315 		/*
316 		 * Buffer output to stdout, stderr while process is grabbed.
317 		 * Prevents infamous deadlocks due to pmap `pgrep xterm` and
318 		 * other variants.
319 		 */
320 		(void) proc_initstdio();
321 
322 		prg_gflags = PGRAB_RETAIN | Fflag;
323 		prr_flags = PRELEASE_RETAIN;
324 
325 		if (sigset(SIGHUP, SIG_IGN) == SIG_DFL)
326 			(void) sigset(SIGHUP, intr);
327 		if (sigset(SIGINT, SIG_IGN) == SIG_DFL)
328 			(void) sigset(SIGINT, intr);
329 		if (sigset(SIGQUIT, SIG_IGN) == SIG_DFL)
330 			(void) sigset(SIGQUIT, intr);
331 		(void) sigset(SIGPIPE, intr);
332 		(void) sigset(SIGTERM, intr);
333 	}
334 
335 	while (argc-- > 0) {
336 		char *arg;
337 		int gcode;
338 		psinfo_t psinfo;
339 		int tries = 0;
340 
341 		if (use_agent_lwp)
342 			(void) proc_flushstdio();
343 
344 		if ((Pr = proc_arg_grab(arg = *argv++, PR_ARG_ANY,
345 		    prg_gflags, &gcode)) == NULL) {
346 			(void) fprintf(stderr, "%s: cannot examine %s: %s\n",
347 			    command, arg, Pgrab_error(gcode));
348 			rc++;
349 			continue;
350 		}
351 
352 		procname = arg;		/* for perr() */
353 
354 		addr_width = (Pstatus(Pr)->pr_dmodel == PR_MODEL_LP64) ? 16 : 8;
355 		size_width = (Pstatus(Pr)->pr_dmodel == PR_MODEL_LP64) ? 11 : 8;
356 		bar = addr_width == 8 ? bar8 : bar16;
357 		(void) memcpy(&psinfo, Ppsinfo(Pr), sizeof (psinfo_t));
358 		proc_unctrl_psinfo(&psinfo);
359 
360 		if (Pstate(Pr) != PS_DEAD) {
361 			(void) snprintf(buf, sizeof (buf),
362 			    "/proc/%d/map", (int)psinfo.pr_pid);
363 			if ((mapfd = open(buf, O_RDONLY)) < 0) {
364 				(void) fprintf(stderr, "%s: cannot "
365 				    "examine %s: lost control of "
366 				    "process\n", command, arg);
367 				rc++;
368 				Prelease(Pr, prr_flags);
369 				continue;
370 			}
371 		} else {
372 			mapfd = -1;
373 		}
374 
375 again:
376 		map_count = 0;
377 
378 		if (Pstate(Pr) == PS_DEAD) {
379 			(void) printf("core '%s' of %d:\t%.70s\n",
380 			    arg, (int)psinfo.pr_pid, psinfo.pr_psargs);
381 
382 			if (rflag || sflag || xflag || Sflag || Lflag) {
383 				(void) printf("  -%c option is not compatible "
384 				    "with core files\n", xflag ? 'x' :
385 				    sflag ? 's' : rflag ? 'r' :
386 				    Lflag ? 'L' : 'S');
387 				Prelease(Pr, prr_flags);
388 				rc++;
389 				continue;
390 			}
391 
392 		} else {
393 			(void) printf("%d:\t%.70s\n",
394 			    (int)psinfo.pr_pid, psinfo.pr_psargs);
395 		}
396 
397 		if (!(Pstatus(Pr)->pr_flags & PR_ISSYS)) {
398 			struct totals t;
399 
400 			/*
401 			 * Since we're grabbing the process readonly, we need
402 			 * to make sure the address space doesn't change during
403 			 * execution.
404 			 */
405 			if (Pstate(Pr) != PS_DEAD) {
406 				if (tries++ == MAX_TRIES) {
407 					Prelease(Pr, prr_flags);
408 					(void) close(mapfd);
409 					(void) fprintf(stderr, "%s: cannot "
410 					    "examine %s: address space is "
411 					    "changing\n", command, arg);
412 					continue;
413 				}
414 
415 				if (fstat64(mapfd, &statbuf) != 0) {
416 					Prelease(Pr, prr_flags);
417 					(void) close(mapfd);
418 					(void) fprintf(stderr, "%s: cannot "
419 					    "examine %s: lost control of "
420 					    "process\n", command, arg);
421 					continue;
422 				}
423 			}
424 
425 			nstacks = psinfo.pr_nlwp * 2;
426 			stacks = calloc(nstacks, sizeof (stacks[0]));
427 			if (stacks != NULL) {
428 				int n = 0;
429 				(void) Plwp_iter(Pr, getstack, &n);
430 				qsort(stacks, nstacks, sizeof (stacks[0]),
431 				    cmpstacks);
432 			}
433 
434 			(void) memset(&t, 0, sizeof (t));
435 
436 			if (Pgetauxval(Pr, AT_BASE) != -1L &&
437 			    Prd_agent(Pr) == NULL) {
438 				(void) fprintf(stderr, "%s: warning: "
439 				    "librtld_db failed to initialize; "
440 				    "shared library information will not be "
441 				    "available\n", command);
442 			}
443 
444 			/*
445 			 * Gather data
446 			 */
447 			if (xflag)
448 				rc += xmapping_iter(Pr, gather_xmap, NULL, 0);
449 			else if (Sflag)
450 				rc += xmapping_iter(Pr, gather_xmap, NULL, 1);
451 			else {
452 				if (rflag)
453 					rc += rmapping_iter(Pr, gather_map,
454 					    NULL);
455 				else if (sflag)
456 					rc += xmapping_iter(Pr, gather_xmap,
457 					    NULL, 0);
458 				else if (lflag)
459 					rc += Pmapping_iter(Pr,
460 					    gather_map, NULL);
461 				else
462 					rc += Pmapping_iter_resolved(Pr,
463 					    gather_map, NULL);
464 			}
465 
466 			/*
467 			 * Ensure mappings are consistent.
468 			 */
469 			if (Pstate(Pr) != PS_DEAD) {
470 				struct stat64 newbuf;
471 
472 				if (fstat64(mapfd, &newbuf) != 0 ||
473 				    memcmp(&newbuf.st_mtim, &statbuf.st_mtim,
474 				    sizeof (newbuf.st_mtim)) != 0) {
475 					if (stacks != NULL) {
476 						free(stacks);
477 						stacks = NULL;
478 					}
479 					goto again;
480 				}
481 			}
482 
483 			/*
484 			 * Display data.
485 			 */
486 			if (xflag) {
487 				(void) printf("%*s%*s%*s%*s%*s "
488 				    "%sMode   Mapped File\n",
489 				    addr_width, "Address",
490 				    size_width, "Kbytes",
491 				    size_width, "RSS",
492 				    size_width, "Anon",
493 				    size_width, "Locked",
494 				    sflag ? "Pgsz " : "");
495 
496 				rc += iter_xmap(sflag ?  look_xmap :
497 				    look_xmap_nopgsz, &t);
498 
499 				(void) printf("%s%s %s %s %s %s\n",
500 				    addr_width == 8 ? "-" : "------",
501 				    bar, bar, bar, bar, bar);
502 
503 				(void) printf("%stotal Kb", addr_width == 16 ?
504 				    "        " : "");
505 
506 				printK(t.total_size, size_width);
507 				printK(t.total_rss, size_width);
508 				printK(t.total_anon, size_width);
509 				printK(t.total_locked, size_width);
510 
511 				(void) printf("\n");
512 
513 			} else if (Sflag) {
514 				(void) printf("%*s%*s%*s Mode"
515 				    " Mapped File\n",
516 				    addr_width, "Address",
517 				    size_width, "Kbytes",
518 				    size_width, "Swap");
519 
520 				rc += iter_xmap(look_xmap_nopgsz, &t);
521 
522 				(void) printf("%s%s %s %s\n",
523 				    addr_width == 8 ? "-" : "------",
524 				    bar, bar, bar);
525 
526 				(void) printf("%stotal Kb", addr_width == 16 ?
527 				    "        " : "");
528 
529 				printK(t.total_size, size_width);
530 				printK(t.total_swap, size_width);
531 
532 				(void) printf("\n");
533 
534 			} else {
535 
536 				if (rflag) {
537 					rc += iter_map(look_map, &t);
538 				} else if (sflag) {
539 					if (Lflag) {
540 						(void) printf("%*s %*s %4s"
541 						    " %-6s %s %s\n",
542 						    addr_width, "Address",
543 						    size_width,
544 						    "Bytes", "Pgsz", "Mode ",
545 						    "Lgrp", "Mapped File");
546 						rc += iter_xmap(look_smap, &t);
547 					} else {
548 						(void) printf("%*s %*s %4s"
549 						    " %-6s %s\n",
550 						    addr_width, "Address",
551 						    size_width,
552 						    "Bytes", "Pgsz", "Mode ",
553 						    "Mapped File");
554 						rc += iter_xmap(look_smap, &t);
555 					}
556 				} else {
557 					rc += iter_map(look_map, &t);
558 				}
559 
560 				(void) printf(" %stotal  %*luK\n",
561 				    addr_width == 16 ?
562 				    "        " : "",
563 				    size_width, t.total_size);
564 			}
565 
566 			if (stacks != NULL) {
567 				free(stacks);
568 				stacks = NULL;
569 			}
570 
571 		}
572 
573 		Prelease(Pr, prr_flags);
574 		if (mapfd != -1)
575 			(void) close(mapfd);
576 	}
577 
578 	if (use_agent_lwp)
579 		(void) proc_finistdio();
580 
581 	return (rc);
582 }
583 
584 static int
rmapping_iter(struct ps_prochandle * Pr,proc_map_f * func,void * cd)585 rmapping_iter(struct ps_prochandle *Pr, proc_map_f *func, void *cd)
586 {
587 	char mapname[PATH_MAX];
588 	int mapfd, nmap, i, rc;
589 	struct stat st;
590 	prmap_t *prmapp, *pmp;
591 	ssize_t n;
592 
593 	(void) snprintf(mapname, sizeof (mapname),
594 	    "/proc/%d/rmap", (int)Pstatus(Pr)->pr_pid);
595 
596 	if ((mapfd = open(mapname, O_RDONLY)) < 0 || fstat(mapfd, &st) != 0) {
597 		if (mapfd >= 0)
598 			(void) close(mapfd);
599 		return (perr(mapname));
600 	}
601 
602 	nmap = st.st_size / sizeof (prmap_t);
603 	prmapp = malloc((nmap + 1) * sizeof (prmap_t));
604 
605 	if ((n = pread(mapfd, prmapp, (nmap + 1) * sizeof (prmap_t), 0L)) < 0) {
606 		(void) close(mapfd);
607 		free(prmapp);
608 		return (perr("read rmap"));
609 	}
610 
611 	(void) close(mapfd);
612 	nmap = n / sizeof (prmap_t);
613 
614 	for (i = 0, pmp = prmapp; i < nmap; i++, pmp++) {
615 		if ((rc = func(cd, pmp, NULL)) != 0) {
616 			free(prmapp);
617 			return (rc);
618 		}
619 	}
620 
621 	free(prmapp);
622 	return (0);
623 }
624 
625 static int
xmapping_iter(struct ps_prochandle * Pr,proc_xmap_f * func,void * cd,int doswap)626 xmapping_iter(struct ps_prochandle *Pr, proc_xmap_f *func, void *cd, int doswap)
627 {
628 	char mapname[PATH_MAX];
629 	int mapfd, nmap, i, rc;
630 	struct stat st;
631 	prxmap_t *prmapp, *pmp;
632 	ssize_t n;
633 
634 	(void) snprintf(mapname, sizeof (mapname),
635 	    "/proc/%d/xmap", (int)Pstatus(Pr)->pr_pid);
636 
637 	if ((mapfd = open(mapname, O_RDONLY)) < 0 || fstat(mapfd, &st) != 0) {
638 		if (mapfd >= 0)
639 			(void) close(mapfd);
640 		return (perr(mapname));
641 	}
642 
643 	nmap = st.st_size / sizeof (prxmap_t);
644 	nmap *= 2;
645 again:
646 	prmapp = malloc((nmap + 1) * sizeof (prxmap_t));
647 
648 	if ((n = pread(mapfd, prmapp, (nmap + 1) * sizeof (prxmap_t), 0)) < 0) {
649 		(void) close(mapfd);
650 		free(prmapp);
651 		return (perr("read xmap"));
652 	}
653 
654 	if (nmap < n / sizeof (prxmap_t)) {
655 		free(prmapp);
656 		nmap *= 2;
657 		goto again;
658 	}
659 
660 	(void) close(mapfd);
661 	nmap = n / sizeof (prxmap_t);
662 
663 	for (i = 0, pmp = prmapp; i < nmap; i++, pmp++) {
664 		if ((rc = func(cd, pmp, NULL, i == nmap - 1, doswap)) != 0) {
665 			free(prmapp);
666 			return (rc);
667 		}
668 	}
669 
670 	/*
671 	 * Mark the last element.
672 	 */
673 	if (map_count > 0)
674 		maps[map_count - 1].md_last = B_TRUE;
675 
676 	free(prmapp);
677 	return (0);
678 }
679 
680 /*ARGSUSED*/
681 static int
look_map(void * data,const prmap_t * pmp,const char * object_name)682 look_map(void *data, const prmap_t *pmp, const char *object_name)
683 {
684 	struct totals *t = data;
685 	const pstatus_t *Psp = Pstatus(Pr);
686 	size_t size;
687 	char mname[PATH_MAX];
688 	char *lname = NULL;
689 	size_t	psz = pmp->pr_pagesize;
690 	uintptr_t vaddr = pmp->pr_vaddr;
691 	uintptr_t segment_end = vaddr + pmp->pr_size;
692 	lgrp_id_t lgrp;
693 	memory_chunk_t mchunk;
694 
695 	/*
696 	 * If the mapping is not anon or not part of the heap, make a name
697 	 * for it.  We don't want to report the heap as a.out's data.
698 	 */
699 	if (!(pmp->pr_mflags & MA_ANON) ||
700 	    segment_end <= Psp->pr_brkbase ||
701 	    pmp->pr_vaddr >= Psp->pr_brkbase + Psp->pr_brksize) {
702 		lname = make_name(Pr, lflag, pmp->pr_vaddr, pmp->pr_mapname,
703 		    mname, sizeof (mname));
704 	}
705 
706 	if (lname == NULL &&
707 	    ((pmp->pr_mflags & MA_ANON) || Pstate(Pr) == PS_DEAD)) {
708 		lname = anon_name(mname, Psp, stacks, nstacks, pmp->pr_vaddr,
709 		    pmp->pr_size, pmp->pr_mflags, pmp->pr_shmid, NULL);
710 	}
711 
712 	/*
713 	 * Adjust the address range if -A is specified.
714 	 */
715 	size = adjust_addr_range(pmp->pr_vaddr, segment_end, psz,
716 	    &vaddr, &segment_end);
717 
718 	if (size == 0)
719 		return (0);
720 
721 	if (!Lflag) {
722 		/*
723 		 * Display the whole mapping
724 		 */
725 		size = ROUNDUP_KB(size);
726 
727 		(void) printf(lname ?
728 		    "%.*lX %*luK %-6s %s\n" :
729 		    "%.*lX %*luK %s\n",
730 		    addr_width, vaddr,
731 		    size_width - 1, size, mflags(pmp->pr_mflags), lname);
732 
733 		t->total_size += size;
734 		return (0);
735 	}
736 
737 	/*
738 	 * We need to display lgroups backing physical memory, so we break the
739 	 * segment into individual pages and coalesce pages with the same lgroup
740 	 * into one "segment".
741 	 */
742 
743 	/*
744 	 * Initialize address descriptions for the mapping.
745 	 */
746 	mem_chunk_init(&mchunk, segment_end, psz);
747 	size = 0;
748 
749 	/*
750 	 * Walk mapping (page by page) and display contiguous ranges of memory
751 	 * allocated to same lgroup.
752 	 */
753 	do {
754 		size_t		size_contig;
755 
756 		/*
757 		 * Get contiguous region of memory starting from vaddr allocated
758 		 * from the same lgroup.
759 		 */
760 		size_contig = get_contiguous_region(&mchunk, vaddr,
761 		    segment_end, pmp->pr_pagesize, &lgrp);
762 
763 		(void) printf(lname ? "%.*lX %*luK %-6s%s %s\n" :
764 		    "%.*lX %*luK %s %s\n",
765 		    addr_width, vaddr,
766 		    size_width - 1, size_contig / KILOBYTE,
767 		    mflags(pmp->pr_mflags),
768 		    lgrp2str(lgrp), lname);
769 
770 		vaddr += size_contig;
771 		size += size_contig;
772 	} while (vaddr < segment_end && !interrupt);
773 
774 	/* Update the total size */
775 	t->total_size += ROUNDUP_KB(size);
776 	return (0);
777 }
778 
779 static void
printK(long value,int width)780 printK(long value, int width)
781 {
782 	if (value == 0)
783 		(void) printf(width == 8 ? "       -" : "          -");
784 	else
785 		(void) printf(" %*lu", width - 1, value);
786 }
787 
788 static const char *
pagesize(const prxmap_t * pmp)789 pagesize(const prxmap_t *pmp)
790 {
791 	int pagesize = pmp->pr_hatpagesize;
792 	static char buf[32];
793 
794 	if (pagesize == 0) {
795 		return ("-"); /* no underlying HAT mapping */
796 	}
797 
798 	if (pagesize >= KILOBYTE && (pagesize % KILOBYTE) == 0) {
799 		if ((pagesize % GIGABYTE) == 0)
800 			(void) snprintf(buf, sizeof (buf), "%dG",
801 			    pagesize / GIGABYTE);
802 		else if ((pagesize % MEGABYTE) == 0)
803 			(void) snprintf(buf, sizeof (buf), "%dM",
804 			    pagesize / MEGABYTE);
805 		else
806 			(void) snprintf(buf, sizeof (buf), "%dK",
807 			    pagesize / KILOBYTE);
808 	} else
809 		(void) snprintf(buf, sizeof (buf), "%db", pagesize);
810 
811 	return (buf);
812 }
813 
814 /*ARGSUSED*/
815 static int
look_smap(void * data,const prxmap_t * pmp,const char * object_name,int last,int doswap)816 look_smap(void *data,
817 	const prxmap_t *pmp,
818 	const char *object_name,
819 	int last, int doswap)
820 {
821 	struct totals *t = data;
822 	const pstatus_t *Psp = Pstatus(Pr);
823 	size_t size;
824 	char mname[PATH_MAX];
825 	char *lname = NULL;
826 	const char *format;
827 	size_t	psz = pmp->pr_pagesize;
828 	uintptr_t vaddr = pmp->pr_vaddr;
829 	uintptr_t segment_end = vaddr + pmp->pr_size;
830 	lgrp_id_t lgrp;
831 	memory_chunk_t mchunk;
832 
833 	/*
834 	 * If the mapping is not anon or not part of the heap, make a name
835 	 * for it.  We don't want to report the heap as a.out's data.
836 	 */
837 	if (!(pmp->pr_mflags & MA_ANON) ||
838 	    pmp->pr_vaddr + pmp->pr_size <= Psp->pr_brkbase ||
839 	    pmp->pr_vaddr >= Psp->pr_brkbase + Psp->pr_brksize) {
840 		lname = make_name(Pr, lflag, pmp->pr_vaddr, pmp->pr_mapname,
841 		    mname, sizeof (mname));
842 	}
843 
844 	if (lname == NULL &&
845 	    ((pmp->pr_mflags & MA_ANON) || Pstate(Pr) == PS_DEAD)) {
846 		lname = anon_name(mname, Psp, stacks, nstacks, pmp->pr_vaddr,
847 		    pmp->pr_size, pmp->pr_mflags, pmp->pr_shmid, NULL);
848 	}
849 
850 	/*
851 	 * Adjust the address range if -A is specified.
852 	 */
853 	size = adjust_addr_range(pmp->pr_vaddr, segment_end, psz,
854 	    &vaddr, &segment_end);
855 
856 	if (size == 0)
857 		return (0);
858 
859 	if (!Lflag) {
860 		/*
861 		 * Display the whole mapping
862 		 */
863 		if (lname != NULL)
864 			format = "%.*lX %*luK %4s %-6s %s\n";
865 		else
866 			format = "%.*lX %*luK %4s %s\n";
867 
868 		size = ROUNDUP_KB(size);
869 
870 		(void) printf(format, addr_width, vaddr, size_width - 1, size,
871 		    pagesize(pmp), mflags(pmp->pr_mflags), lname);
872 
873 		t->total_size += size;
874 		return (0);
875 	}
876 
877 	if (lname != NULL)
878 		format = "%.*lX %*luK %4s %-6s%s %s\n";
879 	else
880 		format = "%.*lX %*luK %4s%s %s\n";
881 
882 	/*
883 	 * We need to display lgroups backing physical memory, so we break the
884 	 * segment into individual pages and coalesce pages with the same lgroup
885 	 * into one "segment".
886 	 */
887 
888 	/*
889 	 * Initialize address descriptions for the mapping.
890 	 */
891 	mem_chunk_init(&mchunk, segment_end, psz);
892 	size = 0;
893 
894 	/*
895 	 * Walk mapping (page by page) and display contiguous ranges of memory
896 	 * allocated to same lgroup.
897 	 */
898 	do {
899 		size_t		size_contig;
900 
901 		/*
902 		 * Get contiguous region of memory starting from vaddr allocated
903 		 * from the same lgroup.
904 		 */
905 		size_contig = get_contiguous_region(&mchunk, vaddr,
906 		    segment_end, pmp->pr_pagesize, &lgrp);
907 
908 		(void) printf(format, addr_width, vaddr,
909 		    size_width - 1, size_contig / KILOBYTE,
910 		    pagesize(pmp), mflags(pmp->pr_mflags),
911 		    lgrp2str(lgrp), lname);
912 
913 		vaddr += size_contig;
914 		size += size_contig;
915 	} while (vaddr < segment_end && !interrupt);
916 
917 	t->total_size += ROUNDUP_KB(size);
918 	return (0);
919 }
920 
921 #define	ANON(x)	((aflag || (((x)->pr_mflags & MA_SHARED) == 0)) ? \
922 	    ((x)->pr_anon) : 0)
923 
924 /*ARGSUSED*/
925 static int
look_xmap(void * data,const prxmap_t * pmp,const char * object_name,int last,int doswap)926 look_xmap(void *data,
927 	const prxmap_t *pmp,
928 	const char *object_name,
929 	int last, int doswap)
930 {
931 	struct totals *t = data;
932 	const pstatus_t *Psp = Pstatus(Pr);
933 	char mname[PATH_MAX];
934 	char *lname = NULL;
935 	char *ln;
936 
937 	/*
938 	 * If the mapping is not anon or not part of the heap, make a name
939 	 * for it.  We don't want to report the heap as a.out's data.
940 	 */
941 	if (!(pmp->pr_mflags & MA_ANON) ||
942 	    pmp->pr_vaddr + pmp->pr_size <= Psp->pr_brkbase ||
943 	    pmp->pr_vaddr >= Psp->pr_brkbase + Psp->pr_brksize) {
944 		lname = make_name(Pr, lflag, pmp->pr_vaddr, pmp->pr_mapname,
945 		    mname, sizeof (mname));
946 	}
947 
948 	if (lname != NULL) {
949 		if ((ln = strrchr(lname, '/')) != NULL)
950 			lname = ln + 1;
951 	} else if ((pmp->pr_mflags & MA_ANON) || Pstate(Pr) == PS_DEAD) {
952 		lname = anon_name(mname, Psp, stacks, nstacks, pmp->pr_vaddr,
953 		    pmp->pr_size, pmp->pr_mflags, pmp->pr_shmid, NULL);
954 	}
955 
956 	(void) printf("%.*lX", addr_width, (ulong_t)pmp->pr_vaddr);
957 
958 	printK(ROUNDUP_KB(pmp->pr_size), size_width);
959 	printK(pmp->pr_rss * (pmp->pr_pagesize / KILOBYTE), size_width);
960 	printK(ANON(pmp) * (pmp->pr_pagesize / KILOBYTE), size_width);
961 	printK(pmp->pr_locked * (pmp->pr_pagesize / KILOBYTE), size_width);
962 	(void) printf(lname ? " %4s %-6s %s\n" : " %4s %s\n",
963 	    pagesize(pmp), mflags(pmp->pr_mflags), lname);
964 
965 	t->total_size += ROUNDUP_KB(pmp->pr_size);
966 	t->total_rss += pmp->pr_rss * (pmp->pr_pagesize / KILOBYTE);
967 	t->total_anon += ANON(pmp) * (pmp->pr_pagesize / KILOBYTE);
968 	t->total_locked += (pmp->pr_locked * (pmp->pr_pagesize / KILOBYTE));
969 
970 	return (0);
971 }
972 
973 /*ARGSUSED*/
974 static int
look_xmap_nopgsz(void * data,const prxmap_t * pmp,const char * object_name,int last,int doswap)975 look_xmap_nopgsz(void *data,
976 	const prxmap_t *pmp,
977 	const char *object_name,
978 	int last, int doswap)
979 {
980 	struct totals *t = data;
981 	const pstatus_t *Psp = Pstatus(Pr);
982 	char mname[PATH_MAX];
983 	char *lname = NULL;
984 	char *ln;
985 	static uintptr_t prev_vaddr;
986 	static size_t prev_size;
987 	static offset_t prev_offset;
988 	static int prev_mflags;
989 	static char *prev_lname;
990 	static char prev_mname[PATH_MAX];
991 	static ulong_t prev_rss;
992 	static ulong_t prev_anon;
993 	static ulong_t prev_locked;
994 	static ulong_t prev_swap;
995 	int merged = 0;
996 	static int first = 1;
997 	ulong_t swap = 0;
998 	int kperpage;
999 
1000 	/*
1001 	 * Calculate swap reservations
1002 	 */
1003 	if (pmp->pr_mflags & MA_SHARED) {
1004 		if (aflag && (pmp->pr_mflags & MA_NORESERVE) == 0) {
1005 			/* Swap reserved for entire non-ism SHM */
1006 			swap = pmp->pr_size / pmp->pr_pagesize;
1007 		}
1008 	} else if (pmp->pr_mflags & MA_NORESERVE) {
1009 		/* Swap reserved on fault for each anon page */
1010 		swap = pmp->pr_anon;
1011 	} else if (pmp->pr_mflags & MA_WRITE) {
1012 		/* Swap reserve for entire writable segment */
1013 		swap = pmp->pr_size / pmp->pr_pagesize;
1014 	}
1015 
1016 	/*
1017 	 * If the mapping is not anon or not part of the heap, make a name
1018 	 * for it.  We don't want to report the heap as a.out's data.
1019 	 */
1020 	if (!(pmp->pr_mflags & MA_ANON) ||
1021 	    pmp->pr_vaddr + pmp->pr_size <= Psp->pr_brkbase ||
1022 	    pmp->pr_vaddr >= Psp->pr_brkbase + Psp->pr_brksize) {
1023 		lname = make_name(Pr, lflag, pmp->pr_vaddr, pmp->pr_mapname,
1024 		    mname, sizeof (mname));
1025 	}
1026 
1027 	if (lname != NULL) {
1028 		if ((ln = strrchr(lname, '/')) != NULL)
1029 			lname = ln + 1;
1030 	} else if ((pmp->pr_mflags & MA_ANON) || Pstate(Pr) == PS_DEAD) {
1031 		lname = anon_name(mname, Psp, stacks, nstacks, pmp->pr_vaddr,
1032 		    pmp->pr_size, pmp->pr_mflags, pmp->pr_shmid, NULL);
1033 	}
1034 
1035 	kperpage = pmp->pr_pagesize / KILOBYTE;
1036 
1037 	t->total_size += ROUNDUP_KB(pmp->pr_size);
1038 	t->total_rss += pmp->pr_rss * kperpage;
1039 	t->total_anon += ANON(pmp) * kperpage;
1040 	t->total_locked += pmp->pr_locked * kperpage;
1041 	t->total_swap += swap * kperpage;
1042 
1043 	if (first == 1) {
1044 		first = 0;
1045 		prev_vaddr = pmp->pr_vaddr;
1046 		prev_size = pmp->pr_size;
1047 		prev_offset = pmp->pr_offset;
1048 		prev_mflags = pmp->pr_mflags;
1049 		if (lname == NULL) {
1050 			prev_lname = NULL;
1051 		} else {
1052 			(void) strcpy(prev_mname, lname);
1053 			prev_lname = prev_mname;
1054 		}
1055 		prev_rss = pmp->pr_rss * kperpage;
1056 		prev_anon = ANON(pmp) * kperpage;
1057 		prev_locked = pmp->pr_locked * kperpage;
1058 		prev_swap = swap * kperpage;
1059 		if (last == 0) {
1060 			return (0);
1061 		}
1062 		merged = 1;
1063 	} else if (prev_vaddr + prev_size == pmp->pr_vaddr &&
1064 	    prev_mflags == pmp->pr_mflags &&
1065 	    ((prev_mflags & MA_ISM) ||
1066 	    prev_offset + prev_size == pmp->pr_offset) &&
1067 	    ((lname == NULL && prev_lname == NULL) ||
1068 	    (lname != NULL && prev_lname != NULL &&
1069 	    strcmp(lname, prev_lname) == 0))) {
1070 		prev_size += pmp->pr_size;
1071 		prev_rss += pmp->pr_rss * kperpage;
1072 		prev_anon += ANON(pmp) * kperpage;
1073 		prev_locked += pmp->pr_locked * kperpage;
1074 		prev_swap += swap * kperpage;
1075 		if (last == 0) {
1076 			return (0);
1077 		}
1078 		merged = 1;
1079 	}
1080 
1081 	(void) printf("%.*lX", addr_width, (ulong_t)prev_vaddr);
1082 	printK(ROUNDUP_KB(prev_size), size_width);
1083 
1084 	if (doswap)
1085 		printK(prev_swap, size_width);
1086 	else {
1087 		printK(prev_rss, size_width);
1088 		printK(prev_anon, size_width);
1089 		printK(prev_locked, size_width);
1090 	}
1091 	(void) printf(prev_lname ? " %-6s %s\n" : "%s\n",
1092 	    mflags(prev_mflags), prev_lname);
1093 
1094 	if (last == 0) {
1095 		prev_vaddr = pmp->pr_vaddr;
1096 		prev_size = pmp->pr_size;
1097 		prev_offset = pmp->pr_offset;
1098 		prev_mflags = pmp->pr_mflags;
1099 		if (lname == NULL) {
1100 			prev_lname = NULL;
1101 		} else {
1102 			(void) strcpy(prev_mname, lname);
1103 			prev_lname = prev_mname;
1104 		}
1105 		prev_rss = pmp->pr_rss * kperpage;
1106 		prev_anon = ANON(pmp) * kperpage;
1107 		prev_locked = pmp->pr_locked * kperpage;
1108 		prev_swap = swap * kperpage;
1109 	} else if (merged == 0) {
1110 		(void) printf("%.*lX", addr_width, (ulong_t)pmp->pr_vaddr);
1111 		printK(ROUNDUP_KB(pmp->pr_size), size_width);
1112 		if (doswap)
1113 			printK(swap * kperpage, size_width);
1114 		else {
1115 			printK(pmp->pr_rss * kperpage, size_width);
1116 			printK(ANON(pmp) * kperpage, size_width);
1117 			printK(pmp->pr_locked * kperpage, size_width);
1118 		}
1119 		(void) printf(lname ? " %-6s %s\n" : " %s\n",
1120 		    mflags(pmp->pr_mflags), lname);
1121 	}
1122 
1123 	if (last != 0)
1124 		first = 1;
1125 
1126 	return (0);
1127 }
1128 
1129 static int
perr(char * s)1130 perr(char *s)
1131 {
1132 	if (s)
1133 		(void) fprintf(stderr, "%s: ", procname);
1134 	else
1135 		s = procname;
1136 	perror(s);
1137 	return (1);
1138 }
1139 
1140 static char *
mflags(uint_t arg)1141 mflags(uint_t arg)
1142 {
1143 	static char code_buf[80];
1144 	char *str = code_buf;
1145 
1146 	/*
1147 	 * rwxsR
1148 	 *
1149 	 * r - segment is readable
1150 	 * w - segment is writable
1151 	 * x - segment is executable
1152 	 * s - segment is shared
1153 	 * R - segment is mapped MAP_NORESERVE
1154 	 *
1155 	 */
1156 	(void) sprintf(str, "%c%c%c%c%c%c",
1157 	    arg & MA_READ ? 'r' : '-',
1158 	    arg & MA_WRITE ? 'w' : '-',
1159 	    arg & MA_EXEC ? 'x' : '-',
1160 	    arg & MA_SHARED ? 's' : '-',
1161 	    arg & MA_NORESERVE ? 'R' : '-',
1162 	    arg & MA_RESERVED1 ? '*' : ' ');
1163 
1164 	return (str);
1165 }
1166 
1167 static mapdata_t *
nextmap(void)1168 nextmap(void)
1169 {
1170 	mapdata_t *newmaps;
1171 	int next;
1172 
1173 	if (map_count == map_alloc) {
1174 		if (map_alloc == 0)
1175 			next = 16;
1176 		else
1177 			next = map_alloc * 2;
1178 
1179 		newmaps = realloc(maps, next * sizeof (mapdata_t));
1180 		if (newmaps == NULL) {
1181 			(void) perr("failed to allocate maps");
1182 			exit(1);
1183 		}
1184 		(void) memset(newmaps + map_alloc, '\0',
1185 		    (next - map_alloc) * sizeof (mapdata_t));
1186 
1187 		map_alloc = next;
1188 		maps = newmaps;
1189 	}
1190 
1191 	return (&maps[map_count++]);
1192 }
1193 
1194 /*ARGSUSED*/
1195 static int
gather_map(void * ignored,const prmap_t * map,const char * objname)1196 gather_map(void *ignored, const prmap_t *map, const char *objname)
1197 {
1198 	mapdata_t *data;
1199 
1200 	/* Skip mappings which are outside the range specified by -A */
1201 	if (!address_in_range(map->pr_vaddr,
1202 	    map->pr_vaddr + map->pr_size, map->pr_pagesize))
1203 		return (0);
1204 
1205 	data = nextmap();
1206 	data->md_map = *map;
1207 	if (data->md_objname != NULL)
1208 		free(data->md_objname);
1209 	data->md_objname = objname ? strdup(objname) : NULL;
1210 
1211 	return (0);
1212 }
1213 
1214 /*ARGSUSED*/
1215 static int
gather_xmap(void * ignored,const prxmap_t * xmap,const char * objname,int last,int doswap)1216 gather_xmap(void *ignored, const prxmap_t *xmap, const char *objname,
1217     int last, int doswap)
1218 {
1219 	mapdata_t *data;
1220 
1221 	/* Skip mappings which are outside the range specified by -A */
1222 	if (!address_in_range(xmap->pr_vaddr,
1223 	    xmap->pr_vaddr + xmap->pr_size, xmap->pr_pagesize))
1224 		return (0);
1225 
1226 	data = nextmap();
1227 	data->md_xmap = *xmap;
1228 	if (data->md_objname != NULL)
1229 		free(data->md_objname);
1230 	data->md_objname = objname ? strdup(objname) : NULL;
1231 	data->md_last = last;
1232 	data->md_doswap = doswap;
1233 
1234 	return (0);
1235 }
1236 
1237 static int
iter_map(proc_map_f * func,void * data)1238 iter_map(proc_map_f *func, void *data)
1239 {
1240 	int i;
1241 	int ret;
1242 
1243 	for (i = 0; i < map_count; i++) {
1244 		if (interrupt)
1245 			break;
1246 		if ((ret = func(data, &maps[i].md_map,
1247 		    maps[i].md_objname)) != 0)
1248 			return (ret);
1249 	}
1250 
1251 	return (0);
1252 }
1253 
1254 static int
iter_xmap(proc_xmap_f * func,void * data)1255 iter_xmap(proc_xmap_f *func, void *data)
1256 {
1257 	int i;
1258 	int ret;
1259 
1260 	for (i = 0; i < map_count; i++) {
1261 		if (interrupt)
1262 			break;
1263 		if ((ret = func(data, &maps[i].md_xmap, maps[i].md_objname,
1264 		    maps[i].md_last, maps[i].md_doswap)) != 0)
1265 			return (ret);
1266 	}
1267 
1268 	return (0);
1269 }
1270 
1271 /*
1272  * Convert lgroup ID to string.
1273  * returns dash when lgroup ID is invalid.
1274  */
1275 static char *
lgrp2str(lgrp_id_t lgrp)1276 lgrp2str(lgrp_id_t lgrp)
1277 {
1278 	static char lgrp_buf[20];
1279 	char *str = lgrp_buf;
1280 
1281 	(void) sprintf(str, lgrp == LGRP_NONE ? "   -" : "%4d", lgrp);
1282 	return (str);
1283 }
1284 
1285 /*
1286  * Parse address range specification for -A option.
1287  * The address range may have the following forms:
1288  *
1289  * address
1290  *	start and end is set to address
1291  * address,
1292  *	start is set to address, end is set to INVALID_ADDRESS
1293  * ,address
1294  *	start is set to 0, end is set to address
1295  * address1,address2
1296  *	start is set to address1, end is set to address2
1297  *
1298  */
1299 static int
parse_addr_range(char * input_str,uintptr_t * start,uintptr_t * end)1300 parse_addr_range(char *input_str, uintptr_t *start, uintptr_t *end)
1301 {
1302 	char *startp = input_str;
1303 	char *endp = strchr(input_str, ',');
1304 	ulong_t	s = (ulong_t)INVALID_ADDRESS;
1305 	ulong_t e = (ulong_t)INVALID_ADDRESS;
1306 
1307 	if (endp != NULL) {
1308 		/*
1309 		 * Comma is present. If there is nothing after comma, the end
1310 		 * remains set at INVALID_ADDRESS. Otherwise it is set to the
1311 		 * value after comma.
1312 		 */
1313 		*endp = '\0';
1314 		endp++;
1315 
1316 		if ((*endp != '\0') && sscanf(endp, "%lx", &e) != 1)
1317 			return (1);
1318 	}
1319 
1320 	if (startp != NULL) {
1321 		/*
1322 		 * Read the start address, if it is specified. If the address is
1323 		 * missing, start will be set to INVALID_ADDRESS.
1324 		 */
1325 		if ((*startp != '\0') && sscanf(startp, "%lx", &s) != 1)
1326 			return (1);
1327 	}
1328 
1329 	/* If there is no comma, end becomes equal to start */
1330 	if (endp == NULL)
1331 		e = s;
1332 
1333 	/*
1334 	 * ,end implies 0..end range
1335 	 */
1336 	if (e != INVALID_ADDRESS && s == INVALID_ADDRESS)
1337 		s = 0;
1338 
1339 	*start = (uintptr_t)s;
1340 	*end = (uintptr_t)e;
1341 
1342 	/* Return error if neither start nor end address were specified */
1343 	return (! (s != INVALID_ADDRESS || e != INVALID_ADDRESS));
1344 }
1345 
1346 /*
1347  * Check whether any portion of [start, end] segment is within the
1348  * [start_addr, end_addr] range.
1349  *
1350  * Return values:
1351  *   0 - address is outside the range
1352  *   1 - address is within the range
1353  */
1354 static int
address_in_range(uintptr_t start,uintptr_t end,size_t psz)1355 address_in_range(uintptr_t start, uintptr_t end, size_t psz)
1356 {
1357 	int rc = 1;
1358 
1359 	/*
1360 	 *  Nothing to do if there is no address range specified with -A
1361 	 */
1362 	if (start_addr != INVALID_ADDRESS || end_addr != INVALID_ADDRESS) {
1363 		/* The segment end is below the range start */
1364 		if ((start_addr != INVALID_ADDRESS) &&
1365 		    (end < P2ALIGN(start_addr, psz)))
1366 			rc = 0;
1367 
1368 		/* The segment start is above the range end */
1369 		if ((end_addr != INVALID_ADDRESS) &&
1370 		    (start > P2ALIGN(end_addr + psz, psz)))
1371 			rc = 0;
1372 	}
1373 	return (rc);
1374 }
1375 
1376 /*
1377  * Returns an intersection of the [start, end] interval and the range specified
1378  * by -A flag [start_addr, end_addr]. Unspecified parts of the address range
1379  * have value INVALID_ADDRESS.
1380  *
1381  * The start_addr address is rounded down to the beginning of page and end_addr
1382  * is rounded up to the end of page.
1383  *
1384  * Returns the size of the resulting interval or zero if the interval is empty
1385  * or invalid.
1386  */
1387 static size_t
adjust_addr_range(uintptr_t start,uintptr_t end,size_t psz,uintptr_t * new_start,uintptr_t * new_end)1388 adjust_addr_range(uintptr_t start, uintptr_t end, size_t psz,
1389     uintptr_t *new_start, uintptr_t *new_end)
1390 {
1391 	uintptr_t from;		/* start_addr rounded down */
1392 	uintptr_t to;		/* end_addr rounded up */
1393 
1394 	/*
1395 	 * Round down the lower address of the range to the beginning of page.
1396 	 */
1397 	if (start_addr == INVALID_ADDRESS) {
1398 		/*
1399 		 * No start_addr specified by -A, the lower part of the interval
1400 		 * does not change.
1401 		 */
1402 		*new_start = start;
1403 	} else {
1404 		from = P2ALIGN(start_addr, psz);
1405 		/*
1406 		 * If end address is outside the range, return an empty
1407 		 * interval
1408 		 */
1409 		if (end <  from) {
1410 			*new_start = *new_end = 0;
1411 			return (0);
1412 		}
1413 		/*
1414 		 * The adjusted start address is the maximum of requested start
1415 		 * and the aligned start_addr of the -A range.
1416 		 */
1417 		*new_start = start < from ? from : start;
1418 	}
1419 
1420 	/*
1421 	 * Round up the higher address of the range to the end of page.
1422 	 */
1423 	if (end_addr == INVALID_ADDRESS) {
1424 		/*
1425 		 * No end_addr specified by -A, the upper part of the interval
1426 		 * does not change.
1427 		 */
1428 		*new_end = end;
1429 	} else {
1430 		/*
1431 		 * If only one address is specified and it is the beginning of a
1432 		 * segment, get information about the whole segment. This
1433 		 * function is called once per segment and the 'end' argument is
1434 		 * always the end of a segment, so just use the 'end' value.
1435 		 */
1436 		to = (end_addr == start_addr && start == start_addr) ?
1437 		    end :
1438 		    P2ALIGN(end_addr + psz, psz);
1439 		/*
1440 		 * If start address is outside the range, return an empty
1441 		 * interval
1442 		 */
1443 		if (start > to) {
1444 			*new_start = *new_end = 0;
1445 			return (0);
1446 		}
1447 		/*
1448 		 * The adjusted end address is the minimum of requested end
1449 		 * and the aligned end_addr of the -A range.
1450 		 */
1451 		*new_end = end > to ? to : end;
1452 	}
1453 
1454 	/*
1455 	 * Make sure that the resulting interval is legal.
1456 	 */
1457 	if (*new_end < *new_start)
1458 			*new_start = *new_end = 0;
1459 
1460 	/* Return the size of the interval */
1461 	return (*new_end - *new_start);
1462 }
1463 
1464 /*
1465  * Initialize memory_info data structure with information about a new segment.
1466  */
1467 static void
mem_chunk_init(memory_chunk_t * chunk,uintptr_t end,size_t psz)1468 mem_chunk_init(memory_chunk_t *chunk, uintptr_t end, size_t psz)
1469 {
1470 	chunk->end_addr = end;
1471 	chunk->page_size = psz;
1472 	chunk->page_index = 0;
1473 	chunk->chunk_start = chunk->chunk_end = 0;
1474 }
1475 
1476 /*
1477  * Create a new chunk of addresses starting from vaddr.
1478  * Pass the whole chunk to pr_meminfo to collect lgroup and page size
1479  * information for each page in the chunk.
1480  */
1481 static void
mem_chunk_get(memory_chunk_t * chunk,uintptr_t vaddr)1482 mem_chunk_get(memory_chunk_t *chunk, uintptr_t vaddr)
1483 {
1484 	page_descr_t	*pdp = chunk->page_info;
1485 	size_t		psz = chunk->page_size;
1486 	uintptr_t	addr = vaddr;
1487 	uint64_t	inaddr[MAX_MEMINFO_CNT];
1488 	uint64_t	outdata[2 * MAX_MEMINFO_CNT];
1489 	uint_t		info[2] = { MEMINFO_VLGRP, MEMINFO_VPAGESIZE };
1490 	uint_t		validity[MAX_MEMINFO_CNT];
1491 	uint64_t	*dataptr = inaddr;
1492 	uint64_t	*outptr = outdata;
1493 	uint_t		*valptr = validity;
1494 	int 		i, j, rc;
1495 
1496 	chunk->chunk_start = vaddr;
1497 	chunk->page_index = 0;	/* reset index for the new chunk */
1498 
1499 	/*
1500 	 * Fill in MAX_MEMINFO_CNT wotrh of pages starting from vaddr. Also,
1501 	 * copy starting address of each page to inaddr array for pr_meminfo.
1502 	 */
1503 	for (i = 0, pdp = chunk->page_info;
1504 	    (i < MAX_MEMINFO_CNT) && (addr <= chunk->end_addr);
1505 	    i++, pdp++, dataptr++, addr += psz) {
1506 		*dataptr = (uint64_t)addr;
1507 		pdp->pd_start = addr;
1508 		pdp->pd_lgrp = LGRP_NONE;
1509 		pdp->pd_valid = 0;
1510 		pdp->pd_pagesize = 0;
1511 	}
1512 
1513 	/* Mark the number of entries in the chunk and the last address */
1514 	chunk->page_count = i;
1515 	chunk->chunk_end = addr - psz;
1516 
1517 	if (interrupt)
1518 		return;
1519 
1520 	/* Call meminfo for all collected addresses */
1521 	rc = pr_meminfo(Pr, inaddr, i, info, 2, outdata, validity);
1522 	if (rc < 0) {
1523 		(void) perr("can not get memory information");
1524 		return;
1525 	}
1526 
1527 	/* Verify validity of each result and fill in the addrs array */
1528 	pdp = chunk->page_info;
1529 	for (j = 0; j < i; j++, pdp++, valptr++, outptr += 2) {
1530 		/* Skip invalid address pointers */
1531 		if ((*valptr & 1) == 0) {
1532 			continue;
1533 		}
1534 
1535 		/* Is lgroup information available? */
1536 		if ((*valptr & 2) != 0) {
1537 			pdp->pd_lgrp = (lgrp_id_t)*outptr;
1538 			pdp->pd_valid = 1;
1539 		}
1540 
1541 		/* Is page size informaion available? */
1542 		if ((*valptr & 4) != 0) {
1543 			pdp->pd_pagesize = *(outptr + 1);
1544 		}
1545 	}
1546 }
1547 
1548 /*
1549  * Starting from address 'vaddr' find the region with pages allocated from the
1550  * same lgroup.
1551  *
1552  * Arguments:
1553  *	mchunk		Initialized memory chunk structure
1554  *	vaddr		Starting address of the region
1555  *	maxaddr		Upper bound of the region
1556  *	pagesize	Default page size to use
1557  *	ret_lgrp	On exit contains the lgroup ID of all pages in the
1558  *			region.
1559  *
1560  * Returns:
1561  *	Size of the contiguous region in bytes
1562  *	The lgroup ID of all pages in the region in ret_lgrp argument.
1563  */
1564 static size_t
get_contiguous_region(memory_chunk_t * mchunk,uintptr_t vaddr,uintptr_t maxaddr,size_t pagesize,lgrp_id_t * ret_lgrp)1565 get_contiguous_region(memory_chunk_t *mchunk, uintptr_t vaddr,
1566     uintptr_t maxaddr, size_t pagesize, lgrp_id_t *ret_lgrp)
1567 {
1568 	size_t		size_contig = 0;
1569 	lgrp_id_t	lgrp;		/* Lgroup of the region start */
1570 	lgrp_id_t	curr_lgrp;	/* Lgroup of the current page */
1571 	size_t		psz = pagesize;	/* Pagesize to use */
1572 
1573 	/* Set both lgroup IDs to the lgroup of the first page */
1574 	curr_lgrp = lgrp = addr_to_lgrp(mchunk, vaddr, &psz);
1575 
1576 	/*
1577 	 * Starting from vaddr, walk page by page until either the end
1578 	 * of the segment is reached or a page is allocated from a different
1579 	 * lgroup. Also stop if interrupted from keyboard.
1580 	 */
1581 	while ((vaddr < maxaddr) && (curr_lgrp == lgrp) && !interrupt) {
1582 		/*
1583 		 * Get lgroup ID and the page size of the current page.
1584 		 */
1585 		curr_lgrp = addr_to_lgrp(mchunk, vaddr, &psz);
1586 		/* If there is no page size information, use the default */
1587 		if (psz == 0)
1588 			psz = pagesize;
1589 
1590 		if (curr_lgrp == lgrp) {
1591 			/*
1592 			 * This page belongs to the contiguous region.
1593 			 * Increase the region size and advance to the new page.
1594 			 */
1595 			size_contig += psz;
1596 			vaddr += psz;
1597 		}
1598 	}
1599 
1600 	/* Return the region lgroup ID and the size */
1601 	*ret_lgrp = lgrp;
1602 	return (size_contig);
1603 }
1604 
1605 /*
1606  * Given a virtual address, return its lgroup and page size. If there is meminfo
1607  * information for an address, use it, otherwise shift the chunk window to the
1608  * vaddr and create a new chunk with known meminfo information.
1609  */
1610 static lgrp_id_t
addr_to_lgrp(memory_chunk_t * chunk,uintptr_t vaddr,size_t * psz)1611 addr_to_lgrp(memory_chunk_t *chunk, uintptr_t vaddr, size_t *psz)
1612 {
1613 	page_descr_t *pdp;
1614 	lgrp_id_t lgrp = LGRP_NONE;
1615 	int i;
1616 
1617 	*psz = chunk->page_size;
1618 
1619 	if (interrupt)
1620 		return (0);
1621 
1622 	/*
1623 	 * Is there information about this address? If not, create a new chunk
1624 	 * starting from vaddr and apply pr_meminfo() to the whole chunk.
1625 	 */
1626 	if (vaddr < chunk->chunk_start || vaddr > chunk->chunk_end) {
1627 		/*
1628 		 * This address is outside the chunk, get the new chunk and
1629 		 * collect meminfo information for it.
1630 		 */
1631 		mem_chunk_get(chunk, vaddr);
1632 	}
1633 
1634 	/*
1635 	 * Find information about the address.
1636 	 */
1637 	pdp = &chunk->page_info[chunk->page_index];
1638 	for (i = chunk->page_index; i < chunk->page_count; i++, pdp++) {
1639 		if (pdp->pd_start == vaddr) {
1640 			if (pdp->pd_valid) {
1641 				lgrp = pdp->pd_lgrp;
1642 				/*
1643 				 * Override page size information if it is
1644 				 * present.
1645 				 */
1646 				if (pdp->pd_pagesize > 0)
1647 					*psz = pdp->pd_pagesize;
1648 			}
1649 			break;
1650 		}
1651 	}
1652 	/*
1653 	 * Remember where we ended - the next search will start here.
1654 	 * We can query for the lgrp for the same address again, so do not
1655 	 * advance index past the current value.
1656 	 */
1657 	chunk->page_index = i;
1658 
1659 	return (lgrp);
1660 }
1661 
1662 /* ARGSUSED */
1663 static void
intr(int sig)1664 intr(int sig)
1665 {
1666 	interrupt = 1;
1667 }
1668