4471 DTrace count() with histogram
[illumos-gate.git] / usr / src / lib / libdtrace / common / dt_consume.c
blob7557e98ae700c66fdf5bde847f8ecd453e6c0768
1 /*
2 * CDDL HEADER START
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.
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.
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]
19 * CDDL HEADER END
22 * Copyright 2009 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
27 * Copyright (c) 2013, Joyent, Inc. All rights reserved.
28 * Copyright (c) 2012 by Delphix. All rights reserved.
31 #include <stdlib.h>
32 #include <strings.h>
33 #include <errno.h>
34 #include <unistd.h>
35 #include <limits.h>
36 #include <assert.h>
37 #include <ctype.h>
38 #include <alloca.h>
39 #include <dt_impl.h>
40 #include <dt_pq.h>
42 #define DT_MASK_LO 0x00000000FFFFFFFFULL
45 * We declare this here because (1) we need it and (2) we want to avoid a
46 * dependency on libm in libdtrace.
48 static long double
49 dt_fabsl(long double x)
51 if (x < 0)
52 return (-x);
54 return (x);
57 static int
58 dt_ndigits(long long val)
60 int rval = 1;
61 long long cmp = 10;
63 if (val < 0) {
64 val = val == INT64_MIN ? INT64_MAX : -val;
65 rval++;
68 while (val > cmp && cmp > 0) {
69 rval++;
70 cmp *= 10;
73 return (rval < 4 ? 4 : rval);
77 * 128-bit arithmetic functions needed to support the stddev() aggregating
78 * action.
80 static int
81 dt_gt_128(uint64_t *a, uint64_t *b)
83 return (a[1] > b[1] || (a[1] == b[1] && a[0] > b[0]));
86 static int
87 dt_ge_128(uint64_t *a, uint64_t *b)
89 return (a[1] > b[1] || (a[1] == b[1] && a[0] >= b[0]));
92 static int
93 dt_le_128(uint64_t *a, uint64_t *b)
95 return (a[1] < b[1] || (a[1] == b[1] && a[0] <= b[0]));
99 * Shift the 128-bit value in a by b. If b is positive, shift left.
100 * If b is negative, shift right.
102 static void
103 dt_shift_128(uint64_t *a, int b)
105 uint64_t mask;
107 if (b == 0)
108 return;
110 if (b < 0) {
111 b = -b;
112 if (b >= 64) {
113 a[0] = a[1] >> (b - 64);
114 a[1] = 0;
115 } else {
116 a[0] >>= b;
117 mask = 1LL << (64 - b);
118 mask -= 1;
119 a[0] |= ((a[1] & mask) << (64 - b));
120 a[1] >>= b;
122 } else {
123 if (b >= 64) {
124 a[1] = a[0] << (b - 64);
125 a[0] = 0;
126 } else {
127 a[1] <<= b;
128 mask = a[0] >> (64 - b);
129 a[1] |= mask;
130 a[0] <<= b;
135 static int
136 dt_nbits_128(uint64_t *a)
138 int nbits = 0;
139 uint64_t tmp[2];
140 uint64_t zero[2] = { 0, 0 };
142 tmp[0] = a[0];
143 tmp[1] = a[1];
145 dt_shift_128(tmp, -1);
146 while (dt_gt_128(tmp, zero)) {
147 dt_shift_128(tmp, -1);
148 nbits++;
151 return (nbits);
154 static void
155 dt_subtract_128(uint64_t *minuend, uint64_t *subtrahend, uint64_t *difference)
157 uint64_t result[2];
159 result[0] = minuend[0] - subtrahend[0];
160 result[1] = minuend[1] - subtrahend[1] -
161 (minuend[0] < subtrahend[0] ? 1 : 0);
163 difference[0] = result[0];
164 difference[1] = result[1];
167 static void
168 dt_add_128(uint64_t *addend1, uint64_t *addend2, uint64_t *sum)
170 uint64_t result[2];
172 result[0] = addend1[0] + addend2[0];
173 result[1] = addend1[1] + addend2[1] +
174 (result[0] < addend1[0] || result[0] < addend2[0] ? 1 : 0);
176 sum[0] = result[0];
177 sum[1] = result[1];
181 * The basic idea is to break the 2 64-bit values into 4 32-bit values,
182 * use native multiplication on those, and then re-combine into the
183 * resulting 128-bit value.
185 * (hi1 << 32 + lo1) * (hi2 << 32 + lo2) =
186 * hi1 * hi2 << 64 +
187 * hi1 * lo2 << 32 +
188 * hi2 * lo1 << 32 +
189 * lo1 * lo2
191 static void
192 dt_multiply_128(uint64_t factor1, uint64_t factor2, uint64_t *product)
194 uint64_t hi1, hi2, lo1, lo2;
195 uint64_t tmp[2];
197 hi1 = factor1 >> 32;
198 hi2 = factor2 >> 32;
200 lo1 = factor1 & DT_MASK_LO;
201 lo2 = factor2 & DT_MASK_LO;
203 product[0] = lo1 * lo2;
204 product[1] = hi1 * hi2;
206 tmp[0] = hi1 * lo2;
207 tmp[1] = 0;
208 dt_shift_128(tmp, 32);
209 dt_add_128(product, tmp, product);
211 tmp[0] = hi2 * lo1;
212 tmp[1] = 0;
213 dt_shift_128(tmp, 32);
214 dt_add_128(product, tmp, product);
218 * This is long-hand division.
220 * We initialize subtrahend by shifting divisor left as far as possible. We
221 * loop, comparing subtrahend to dividend: if subtrahend is smaller, we
222 * subtract and set the appropriate bit in the result. We then shift
223 * subtrahend right by one bit for the next comparison.
225 static void
226 dt_divide_128(uint64_t *dividend, uint64_t divisor, uint64_t *quotient)
228 uint64_t result[2] = { 0, 0 };
229 uint64_t remainder[2];
230 uint64_t subtrahend[2];
231 uint64_t divisor_128[2];
232 uint64_t mask[2] = { 1, 0 };
233 int log = 0;
235 assert(divisor != 0);
237 divisor_128[0] = divisor;
238 divisor_128[1] = 0;
240 remainder[0] = dividend[0];
241 remainder[1] = dividend[1];
243 subtrahend[0] = divisor;
244 subtrahend[1] = 0;
246 while (divisor > 0) {
247 log++;
248 divisor >>= 1;
251 dt_shift_128(subtrahend, 128 - log);
252 dt_shift_128(mask, 128 - log);
254 while (dt_ge_128(remainder, divisor_128)) {
255 if (dt_ge_128(remainder, subtrahend)) {
256 dt_subtract_128(remainder, subtrahend, remainder);
257 result[0] |= mask[0];
258 result[1] |= mask[1];
261 dt_shift_128(subtrahend, -1);
262 dt_shift_128(mask, -1);
265 quotient[0] = result[0];
266 quotient[1] = result[1];
270 * This is the long-hand method of calculating a square root.
271 * The algorithm is as follows:
273 * 1. Group the digits by 2 from the right.
274 * 2. Over the leftmost group, find the largest single-digit number
275 * whose square is less than that group.
276 * 3. Subtract the result of the previous step (2 or 4, depending) and
277 * bring down the next two-digit group.
278 * 4. For the result R we have so far, find the largest single-digit number
279 * x such that 2 * R * 10 * x + x^2 is less than the result from step 3.
280 * (Note that this is doubling R and performing a decimal left-shift by 1
281 * and searching for the appropriate decimal to fill the one's place.)
282 * The value x is the next digit in the square root.
283 * Repeat steps 3 and 4 until the desired precision is reached. (We're
284 * dealing with integers, so the above is sufficient.)
286 * In decimal, the square root of 582,734 would be calculated as so:
288 * __7__6__3
289 * | 58 27 34
290 * -49 (7^2 == 49 => 7 is the first digit in the square root)
291 * --
292 * 9 27 (Subtract and bring down the next group.)
293 * 146 8 76 (2 * 7 * 10 * 6 + 6^2 == 876 => 6 is the next digit in
294 * ----- the square root)
295 * 51 34 (Subtract and bring down the next group.)
296 * 1523 45 69 (2 * 76 * 10 * 3 + 3^2 == 4569 => 3 is the next digit in
297 * ----- the square root)
298 * 5 65 (remainder)
300 * The above algorithm applies similarly in binary, but note that the
301 * only possible non-zero value for x in step 4 is 1, so step 4 becomes a
302 * simple decision: is 2 * R * 2 * 1 + 1^2 (aka R << 2 + 1) less than the
303 * preceding difference?
305 * In binary, the square root of 11011011 would be calculated as so:
307 * __1__1__1__0
308 * | 11 01 10 11
309 * 01 (0 << 2 + 1 == 1 < 11 => this bit is 1)
310 * --
311 * 10 01 10 11
312 * 101 1 01 (1 << 2 + 1 == 101 < 1001 => next bit is 1)
313 * -----
314 * 1 00 10 11
315 * 1101 11 01 (11 << 2 + 1 == 1101 < 10010 => next bit is 1)
316 * -------
317 * 1 01 11
318 * 11101 1 11 01 (111 << 2 + 1 == 11101 > 10111 => last bit is 0)
321 static uint64_t
322 dt_sqrt_128(uint64_t *square)
324 uint64_t result[2] = { 0, 0 };
325 uint64_t diff[2] = { 0, 0 };
326 uint64_t one[2] = { 1, 0 };
327 uint64_t next_pair[2];
328 uint64_t next_try[2];
329 uint64_t bit_pairs, pair_shift;
330 int i;
332 bit_pairs = dt_nbits_128(square) / 2;
333 pair_shift = bit_pairs * 2;
335 for (i = 0; i <= bit_pairs; i++) {
337 * Bring down the next pair of bits.
339 next_pair[0] = square[0];
340 next_pair[1] = square[1];
341 dt_shift_128(next_pair, -pair_shift);
342 next_pair[0] &= 0x3;
343 next_pair[1] = 0;
345 dt_shift_128(diff, 2);
346 dt_add_128(diff, next_pair, diff);
349 * next_try = R << 2 + 1
351 next_try[0] = result[0];
352 next_try[1] = result[1];
353 dt_shift_128(next_try, 2);
354 dt_add_128(next_try, one, next_try);
356 if (dt_le_128(next_try, diff)) {
357 dt_subtract_128(diff, next_try, diff);
358 dt_shift_128(result, 1);
359 dt_add_128(result, one, result);
360 } else {
361 dt_shift_128(result, 1);
364 pair_shift -= 2;
367 assert(result[1] == 0);
369 return (result[0]);
372 uint64_t
373 dt_stddev(uint64_t *data, uint64_t normal)
375 uint64_t avg_of_squares[2];
376 uint64_t square_of_avg[2];
377 int64_t norm_avg;
378 uint64_t diff[2];
381 * The standard approximation for standard deviation is
382 * sqrt(average(x**2) - average(x)**2), i.e. the square root
383 * of the average of the squares minus the square of the average.
385 dt_divide_128(data + 2, normal, avg_of_squares);
386 dt_divide_128(avg_of_squares, data[0], avg_of_squares);
388 norm_avg = (int64_t)data[1] / (int64_t)normal / (int64_t)data[0];
390 if (norm_avg < 0)
391 norm_avg = -norm_avg;
393 dt_multiply_128((uint64_t)norm_avg, (uint64_t)norm_avg, square_of_avg);
395 dt_subtract_128(avg_of_squares, square_of_avg, diff);
397 return (dt_sqrt_128(diff));
400 static int
401 dt_flowindent(dtrace_hdl_t *dtp, dtrace_probedata_t *data, dtrace_epid_t last,
402 dtrace_bufdesc_t *buf, size_t offs)
404 dtrace_probedesc_t *pd = data->dtpda_pdesc, *npd;
405 dtrace_eprobedesc_t *epd = data->dtpda_edesc, *nepd;
406 char *p = pd->dtpd_provider, *n = pd->dtpd_name, *sub;
407 dtrace_flowkind_t flow = DTRACEFLOW_NONE;
408 const char *str = NULL;
409 static const char *e_str[2] = { " -> ", " => " };
410 static const char *r_str[2] = { " <- ", " <= " };
411 static const char *ent = "entry", *ret = "return";
412 static int entlen = 0, retlen = 0;
413 dtrace_epid_t next, id = epd->dtepd_epid;
414 int rval;
416 if (entlen == 0) {
417 assert(retlen == 0);
418 entlen = strlen(ent);
419 retlen = strlen(ret);
423 * If the name of the probe is "entry" or ends with "-entry", we
424 * treat it as an entry; if it is "return" or ends with "-return",
425 * we treat it as a return. (This allows application-provided probes
426 * like "method-entry" or "function-entry" to participate in flow
427 * indentation -- without accidentally misinterpreting popular probe
428 * names like "carpentry", "gentry" or "Coventry".)
430 if ((sub = strstr(n, ent)) != NULL && sub[entlen] == '\0' &&
431 (sub == n || sub[-1] == '-')) {
432 flow = DTRACEFLOW_ENTRY;
433 str = e_str[strcmp(p, "syscall") == 0];
434 } else if ((sub = strstr(n, ret)) != NULL && sub[retlen] == '\0' &&
435 (sub == n || sub[-1] == '-')) {
436 flow = DTRACEFLOW_RETURN;
437 str = r_str[strcmp(p, "syscall") == 0];
441 * If we're going to indent this, we need to check the ID of our last
442 * call. If we're looking at the same probe ID but a different EPID,
443 * we _don't_ want to indent. (Yes, there are some minor holes in
444 * this scheme -- it's a heuristic.)
446 if (flow == DTRACEFLOW_ENTRY) {
447 if ((last != DTRACE_EPIDNONE && id != last &&
448 pd->dtpd_id == dtp->dt_pdesc[last]->dtpd_id))
449 flow = DTRACEFLOW_NONE;
453 * If we're going to unindent this, it's more difficult to see if
454 * we don't actually want to unindent it -- we need to look at the
455 * _next_ EPID.
457 if (flow == DTRACEFLOW_RETURN) {
458 offs += epd->dtepd_size;
460 do {
461 if (offs >= buf->dtbd_size)
462 goto out;
464 next = *(uint32_t *)((uintptr_t)buf->dtbd_data + offs);
466 if (next == DTRACE_EPIDNONE)
467 offs += sizeof (id);
468 } while (next == DTRACE_EPIDNONE);
470 if ((rval = dt_epid_lookup(dtp, next, &nepd, &npd)) != 0)
471 return (rval);
473 if (next != id && npd->dtpd_id == pd->dtpd_id)
474 flow = DTRACEFLOW_NONE;
477 out:
478 if (flow == DTRACEFLOW_ENTRY || flow == DTRACEFLOW_RETURN) {
479 data->dtpda_prefix = str;
480 } else {
481 data->dtpda_prefix = "| ";
484 if (flow == DTRACEFLOW_RETURN && data->dtpda_indent > 0)
485 data->dtpda_indent -= 2;
487 data->dtpda_flow = flow;
489 return (0);
492 static int
493 dt_nullprobe()
495 return (DTRACE_CONSUME_THIS);
498 static int
499 dt_nullrec()
501 return (DTRACE_CONSUME_NEXT);
504 static void
505 dt_quantize_total(dtrace_hdl_t *dtp, int64_t datum, long double *total)
507 long double val = dt_fabsl((long double)datum);
509 if (dtp->dt_options[DTRACEOPT_AGGZOOM] == DTRACEOPT_UNSET) {
510 *total += val;
511 return;
515 * If we're zooming in on an aggregation, we want the height of the
516 * highest value to be approximately 95% of total bar height -- so we
517 * adjust up by the reciprocal of DTRACE_AGGZOOM_MAX when comparing to
518 * our highest value.
520 val *= 1 / DTRACE_AGGZOOM_MAX;
522 if (*total < val)
523 *total = val;
526 static int
527 dt_print_quanthdr(dtrace_hdl_t *dtp, FILE *fp, int width)
529 return (dt_printf(dtp, fp, "\n%*s %41s %-9s\n",
530 width ? width : 16, width ? "key" : "value",
531 "------------- Distribution -------------", "count"));
534 static int
535 dt_print_quanthdr_packed(dtrace_hdl_t *dtp, FILE *fp, int width,
536 const dtrace_aggdata_t *aggdata, dtrace_actkind_t action)
538 int min = aggdata->dtada_minbin, max = aggdata->dtada_maxbin;
539 int minwidth, maxwidth, i;
541 assert(action == DTRACEAGG_QUANTIZE || action == DTRACEAGG_LQUANTIZE);
543 if (action == DTRACEAGG_QUANTIZE) {
544 if (min != 0 && min != DTRACE_QUANTIZE_ZEROBUCKET)
545 min--;
547 if (max < DTRACE_QUANTIZE_NBUCKETS - 1)
548 max++;
550 minwidth = dt_ndigits(DTRACE_QUANTIZE_BUCKETVAL(min));
551 maxwidth = dt_ndigits(DTRACE_QUANTIZE_BUCKETVAL(max));
552 } else {
553 maxwidth = 8;
554 minwidth = maxwidth - 1;
555 max++;
558 if (dt_printf(dtp, fp, "\n%*s %*s .",
559 width, width > 0 ? "key" : "", minwidth, "min") < 0)
560 return (-1);
562 for (i = min; i <= max; i++) {
563 if (dt_printf(dtp, fp, "-") < 0)
564 return (-1);
567 return (dt_printf(dtp, fp, ". %*s | count\n", -maxwidth, "max"));
571 * We use a subset of the Unicode Block Elements (U+2588 through U+258F,
572 * inclusive) to represent aggregations via UTF-8 -- which are expressed via
573 * 3-byte UTF-8 sequences.
575 #define DTRACE_AGGUTF8_FULL 0x2588
576 #define DTRACE_AGGUTF8_BASE 0x258f
577 #define DTRACE_AGGUTF8_LEVELS 8
579 #define DTRACE_AGGUTF8_BYTE0(val) (0xe0 | ((val) >> 12))
580 #define DTRACE_AGGUTF8_BYTE1(val) (0x80 | (((val) >> 6) & 0x3f))
581 #define DTRACE_AGGUTF8_BYTE2(val) (0x80 | ((val) & 0x3f))
583 static int
584 dt_print_quantline_utf8(dtrace_hdl_t *dtp, FILE *fp, int64_t val,
585 uint64_t normal, long double total)
587 uint_t len = 40, i, whole, partial;
588 long double f = (dt_fabsl((long double)val) * len) / total;
589 const char *spaces = " ";
591 whole = (uint_t)f;
592 partial = (uint_t)((f - (long double)(uint_t)f) *
593 (long double)DTRACE_AGGUTF8_LEVELS);
595 if (dt_printf(dtp, fp, "|") < 0)
596 return (-1);
598 for (i = 0; i < whole; i++) {
599 if (dt_printf(dtp, fp, "%c%c%c",
600 DTRACE_AGGUTF8_BYTE0(DTRACE_AGGUTF8_FULL),
601 DTRACE_AGGUTF8_BYTE1(DTRACE_AGGUTF8_FULL),
602 DTRACE_AGGUTF8_BYTE2(DTRACE_AGGUTF8_FULL)) < 0)
603 return (-1);
606 if (partial != 0) {
607 partial = DTRACE_AGGUTF8_BASE - (partial - 1);
609 if (dt_printf(dtp, fp, "%c%c%c",
610 DTRACE_AGGUTF8_BYTE0(partial),
611 DTRACE_AGGUTF8_BYTE1(partial),
612 DTRACE_AGGUTF8_BYTE2(partial)) < 0)
613 return (-1);
615 i++;
618 return (dt_printf(dtp, fp, "%s %-9lld\n", spaces + i,
619 (long long)val / normal));
622 static int
623 dt_print_quantline(dtrace_hdl_t *dtp, FILE *fp, int64_t val,
624 uint64_t normal, long double total, char positives, char negatives)
626 long double f;
627 uint_t depth, len = 40;
629 const char *ats = "@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@";
630 const char *spaces = " ";
632 assert(strlen(ats) == len && strlen(spaces) == len);
633 assert(!(total == 0 && (positives || negatives)));
634 assert(!(val < 0 && !negatives));
635 assert(!(val > 0 && !positives));
636 assert(!(val != 0 && total == 0));
638 if (!negatives) {
639 if (positives) {
640 if (dtp->dt_encoding == DT_ENCODING_UTF8) {
641 return (dt_print_quantline_utf8(dtp, fp, val,
642 normal, total));
645 f = (dt_fabsl((long double)val) * len) / total;
646 depth = (uint_t)(f + 0.5);
647 } else {
648 depth = 0;
651 return (dt_printf(dtp, fp, "|%s%s %-9lld\n", ats + len - depth,
652 spaces + depth, (long long)val / normal));
655 if (!positives) {
656 f = (dt_fabsl((long double)val) * len) / total;
657 depth = (uint_t)(f + 0.5);
659 return (dt_printf(dtp, fp, "%s%s| %-9lld\n", spaces + depth,
660 ats + len - depth, (long long)val / normal));
664 * If we're here, we have both positive and negative bucket values.
665 * To express this graphically, we're going to generate both positive
666 * and negative bars separated by a centerline. These bars are half
667 * the size of normal quantize()/lquantize() bars, so we divide the
668 * length in half before calculating the bar length.
670 len /= 2;
671 ats = &ats[len];
672 spaces = &spaces[len];
674 f = (dt_fabsl((long double)val) * len) / total;
675 depth = (uint_t)(f + 0.5);
677 if (val <= 0) {
678 return (dt_printf(dtp, fp, "%s%s|%*s %-9lld\n", spaces + depth,
679 ats + len - depth, len, "", (long long)val / normal));
680 } else {
681 return (dt_printf(dtp, fp, "%20s|%s%s %-9lld\n", "",
682 ats + len - depth, spaces + depth,
683 (long long)val / normal));
688 * As with UTF-8 printing of aggregations, we use a subset of the Unicode
689 * Block Elements (U+2581 through U+2588, inclusive) to represent our packed
690 * aggregation.
692 #define DTRACE_AGGPACK_BASE 0x2581
693 #define DTRACE_AGGPACK_LEVELS 8
695 static int
696 dt_print_packed(dtrace_hdl_t *dtp, FILE *fp,
697 long double datum, long double total)
699 static boolean_t utf8_checked = B_FALSE;
700 static boolean_t utf8;
701 char *ascii = "__xxxxXX";
702 char *neg = "vvvvVV";
703 unsigned int len;
704 long double val;
706 if (!utf8_checked) {
707 char *term;
710 * We want to determine if we can reasonably emit UTF-8 for our
711 * packed aggregation. To do this, we will check for terminals
712 * that are known to be primitive to emit UTF-8 on these.
714 utf8_checked = B_TRUE;
716 if (dtp->dt_encoding == DT_ENCODING_ASCII) {
717 utf8 = B_FALSE;
718 } else if (dtp->dt_encoding == DT_ENCODING_UTF8) {
719 utf8 = B_TRUE;
720 } else if ((term = getenv("TERM")) != NULL &&
721 (strcmp(term, "sun") == 0 ||
722 strcmp(term, "sun-color") == 0) ||
723 strcmp(term, "dumb") == 0) {
724 utf8 = B_FALSE;
725 } else {
726 utf8 = B_TRUE;
730 if (datum == 0)
731 return (dt_printf(dtp, fp, " "));
733 if (datum < 0) {
734 len = strlen(neg);
735 val = dt_fabsl(datum * (len - 1)) / total;
736 return (dt_printf(dtp, fp, "%c", neg[(uint_t)(val + 0.5)]));
739 if (utf8) {
740 int block = DTRACE_AGGPACK_BASE + (unsigned int)(((datum *
741 (DTRACE_AGGPACK_LEVELS - 1)) / total) + 0.5);
743 return (dt_printf(dtp, fp, "%c%c%c",
744 DTRACE_AGGUTF8_BYTE0(block),
745 DTRACE_AGGUTF8_BYTE1(block),
746 DTRACE_AGGUTF8_BYTE2(block)));
749 len = strlen(ascii);
750 val = (datum * (len - 1)) / total;
751 return (dt_printf(dtp, fp, "%c", ascii[(uint_t)(val + 0.5)]));
755 dt_print_quantize(dtrace_hdl_t *dtp, FILE *fp, const void *addr,
756 size_t size, uint64_t normal)
758 const int64_t *data = addr;
759 int i, first_bin = 0, last_bin = DTRACE_QUANTIZE_NBUCKETS - 1;
760 long double total = 0;
761 char positives = 0, negatives = 0;
763 if (size != DTRACE_QUANTIZE_NBUCKETS * sizeof (uint64_t))
764 return (dt_set_errno(dtp, EDT_DMISMATCH));
766 while (first_bin < DTRACE_QUANTIZE_NBUCKETS - 1 && data[first_bin] == 0)
767 first_bin++;
769 if (first_bin == DTRACE_QUANTIZE_NBUCKETS - 1) {
771 * There isn't any data. This is possible if the aggregation
772 * has been clear()'d or if negative increment values have been
773 * used. Regardless, we'll print the buckets around 0.
775 first_bin = DTRACE_QUANTIZE_ZEROBUCKET - 1;
776 last_bin = DTRACE_QUANTIZE_ZEROBUCKET + 1;
777 } else {
778 if (first_bin > 0)
779 first_bin--;
781 while (last_bin > 0 && data[last_bin] == 0)
782 last_bin--;
784 if (last_bin < DTRACE_QUANTIZE_NBUCKETS - 1)
785 last_bin++;
788 for (i = first_bin; i <= last_bin; i++) {
789 positives |= (data[i] > 0);
790 negatives |= (data[i] < 0);
791 dt_quantize_total(dtp, data[i], &total);
794 if (dt_print_quanthdr(dtp, fp, 0) < 0)
795 return (-1);
797 for (i = first_bin; i <= last_bin; i++) {
798 if (dt_printf(dtp, fp, "%16lld ",
799 (long long)DTRACE_QUANTIZE_BUCKETVAL(i)) < 0)
800 return (-1);
802 if (dt_print_quantline(dtp, fp, data[i], normal, total,
803 positives, negatives) < 0)
804 return (-1);
807 return (0);
811 dt_print_quantize_packed(dtrace_hdl_t *dtp, FILE *fp, const void *addr,
812 size_t size, const dtrace_aggdata_t *aggdata)
814 const int64_t *data = addr;
815 long double total = 0, count = 0;
816 int min = aggdata->dtada_minbin, max = aggdata->dtada_maxbin, i;
817 int64_t minval, maxval;
819 if (size != DTRACE_QUANTIZE_NBUCKETS * sizeof (uint64_t))
820 return (dt_set_errno(dtp, EDT_DMISMATCH));
822 if (min != 0 && min != DTRACE_QUANTIZE_ZEROBUCKET)
823 min--;
825 if (max < DTRACE_QUANTIZE_NBUCKETS - 1)
826 max++;
828 minval = DTRACE_QUANTIZE_BUCKETVAL(min);
829 maxval = DTRACE_QUANTIZE_BUCKETVAL(max);
831 if (dt_printf(dtp, fp, " %*lld :", dt_ndigits(minval),
832 (long long)minval) < 0)
833 return (-1);
835 for (i = min; i <= max; i++) {
836 dt_quantize_total(dtp, data[i], &total);
837 count += data[i];
840 for (i = min; i <= max; i++) {
841 if (dt_print_packed(dtp, fp, data[i], total) < 0)
842 return (-1);
845 if (dt_printf(dtp, fp, ": %*lld | %lld\n",
846 -dt_ndigits(maxval), (long long)maxval, (long long)count) < 0)
847 return (-1);
849 return (0);
853 dt_print_lquantize(dtrace_hdl_t *dtp, FILE *fp, const void *addr,
854 size_t size, uint64_t normal)
856 const int64_t *data = addr;
857 int i, first_bin, last_bin, base;
858 uint64_t arg;
859 long double total = 0;
860 uint16_t step, levels;
861 char positives = 0, negatives = 0;
863 if (size < sizeof (uint64_t))
864 return (dt_set_errno(dtp, EDT_DMISMATCH));
866 arg = *data++;
867 size -= sizeof (uint64_t);
869 base = DTRACE_LQUANTIZE_BASE(arg);
870 step = DTRACE_LQUANTIZE_STEP(arg);
871 levels = DTRACE_LQUANTIZE_LEVELS(arg);
873 first_bin = 0;
874 last_bin = levels + 1;
876 if (size != sizeof (uint64_t) * (levels + 2))
877 return (dt_set_errno(dtp, EDT_DMISMATCH));
879 while (first_bin <= levels + 1 && data[first_bin] == 0)
880 first_bin++;
882 if (first_bin > levels + 1) {
883 first_bin = 0;
884 last_bin = 2;
885 } else {
886 if (first_bin > 0)
887 first_bin--;
889 while (last_bin > 0 && data[last_bin] == 0)
890 last_bin--;
892 if (last_bin < levels + 1)
893 last_bin++;
896 for (i = first_bin; i <= last_bin; i++) {
897 positives |= (data[i] > 0);
898 negatives |= (data[i] < 0);
899 dt_quantize_total(dtp, data[i], &total);
902 if (dt_printf(dtp, fp, "\n%16s %41s %-9s\n", "value",
903 "------------- Distribution -------------", "count") < 0)
904 return (-1);
906 for (i = first_bin; i <= last_bin; i++) {
907 char c[32];
908 int err;
910 if (i == 0) {
911 (void) snprintf(c, sizeof (c), "< %d", base);
912 err = dt_printf(dtp, fp, "%16s ", c);
913 } else if (i == levels + 1) {
914 (void) snprintf(c, sizeof (c), ">= %d",
915 base + (levels * step));
916 err = dt_printf(dtp, fp, "%16s ", c);
917 } else {
918 err = dt_printf(dtp, fp, "%16d ",
919 base + (i - 1) * step);
922 if (err < 0 || dt_print_quantline(dtp, fp, data[i], normal,
923 total, positives, negatives) < 0)
924 return (-1);
927 return (0);
930 /*ARGSUSED*/
932 dt_print_lquantize_packed(dtrace_hdl_t *dtp, FILE *fp, const void *addr,
933 size_t size, const dtrace_aggdata_t *aggdata)
935 const int64_t *data = addr;
936 long double total = 0, count = 0;
937 int min, max, base, err;
938 uint64_t arg;
939 uint16_t step, levels;
940 char c[32];
941 unsigned int i;
943 if (size < sizeof (uint64_t))
944 return (dt_set_errno(dtp, EDT_DMISMATCH));
946 arg = *data++;
947 size -= sizeof (uint64_t);
949 base = DTRACE_LQUANTIZE_BASE(arg);
950 step = DTRACE_LQUANTIZE_STEP(arg);
951 levels = DTRACE_LQUANTIZE_LEVELS(arg);
953 if (size != sizeof (uint64_t) * (levels + 2))
954 return (dt_set_errno(dtp, EDT_DMISMATCH));
956 min = 0;
957 max = levels + 1;
959 if (min == 0) {
960 (void) snprintf(c, sizeof (c), "< %d", base);
961 err = dt_printf(dtp, fp, "%8s :", c);
962 } else {
963 err = dt_printf(dtp, fp, "%8d :", base + (min - 1) * step);
966 if (err < 0)
967 return (-1);
969 for (i = min; i <= max; i++) {
970 dt_quantize_total(dtp, data[i], &total);
971 count += data[i];
974 for (i = min; i <= max; i++) {
975 if (dt_print_packed(dtp, fp, data[i], total) < 0)
976 return (-1);
979 (void) snprintf(c, sizeof (c), ">= %d", base + (levels * step));
980 return (dt_printf(dtp, fp, ": %-8s | %lld\n", c, (long long)count));
984 dt_print_llquantize(dtrace_hdl_t *dtp, FILE *fp, const void *addr,
985 size_t size, uint64_t normal)
987 int i, first_bin, last_bin, bin = 1, order, levels;
988 uint16_t factor, low, high, nsteps;
989 const int64_t *data = addr;
990 int64_t value = 1, next, step;
991 char positives = 0, negatives = 0;
992 long double total = 0;
993 uint64_t arg;
994 char c[32];
996 if (size < sizeof (uint64_t))
997 return (dt_set_errno(dtp, EDT_DMISMATCH));
999 arg = *data++;
1000 size -= sizeof (uint64_t);
1002 factor = DTRACE_LLQUANTIZE_FACTOR(arg);
1003 low = DTRACE_LLQUANTIZE_LOW(arg);
1004 high = DTRACE_LLQUANTIZE_HIGH(arg);
1005 nsteps = DTRACE_LLQUANTIZE_NSTEP(arg);
1008 * We don't expect to be handed invalid llquantize() parameters here,
1009 * but sanity check them (to a degree) nonetheless.
1011 if (size > INT32_MAX || factor < 2 || low >= high ||
1012 nsteps == 0 || factor > nsteps)
1013 return (dt_set_errno(dtp, EDT_DMISMATCH));
1015 levels = (int)size / sizeof (uint64_t);
1017 first_bin = 0;
1018 last_bin = levels - 1;
1020 while (first_bin < levels && data[first_bin] == 0)
1021 first_bin++;
1023 if (first_bin == levels) {
1024 first_bin = 0;
1025 last_bin = 1;
1026 } else {
1027 if (first_bin > 0)
1028 first_bin--;
1030 while (last_bin > 0 && data[last_bin] == 0)
1031 last_bin--;
1033 if (last_bin < levels - 1)
1034 last_bin++;
1037 for (i = first_bin; i <= last_bin; i++) {
1038 positives |= (data[i] > 0);
1039 negatives |= (data[i] < 0);
1040 dt_quantize_total(dtp, data[i], &total);
1043 if (dt_printf(dtp, fp, "\n%16s %41s %-9s\n", "value",
1044 "------------- Distribution -------------", "count") < 0)
1045 return (-1);
1047 for (order = 0; order < low; order++)
1048 value *= factor;
1050 next = value * factor;
1051 step = next > nsteps ? next / nsteps : 1;
1053 if (first_bin == 0) {
1054 (void) snprintf(c, sizeof (c), "< %lld", value);
1056 if (dt_printf(dtp, fp, "%16s ", c) < 0)
1057 return (-1);
1059 if (dt_print_quantline(dtp, fp, data[0], normal,
1060 total, positives, negatives) < 0)
1061 return (-1);
1064 while (order <= high) {
1065 if (bin >= first_bin && bin <= last_bin) {
1066 if (dt_printf(dtp, fp, "%16lld ", (long long)value) < 0)
1067 return (-1);
1069 if (dt_print_quantline(dtp, fp, data[bin],
1070 normal, total, positives, negatives) < 0)
1071 return (-1);
1074 assert(value < next);
1075 bin++;
1077 if ((value += step) != next)
1078 continue;
1080 next = value * factor;
1081 step = next > nsteps ? next / nsteps : 1;
1082 order++;
1085 if (last_bin < bin)
1086 return (0);
1088 assert(last_bin == bin);
1089 (void) snprintf(c, sizeof (c), ">= %lld", value);
1091 if (dt_printf(dtp, fp, "%16s ", c) < 0)
1092 return (-1);
1094 return (dt_print_quantline(dtp, fp, data[bin], normal,
1095 total, positives, negatives));
1098 /*ARGSUSED*/
1099 static int
1100 dt_print_average(dtrace_hdl_t *dtp, FILE *fp, caddr_t addr,
1101 size_t size, uint64_t normal)
1103 /* LINTED - alignment */
1104 int64_t *data = (int64_t *)addr;
1106 return (dt_printf(dtp, fp, " %16lld", data[0] ?
1107 (long long)(data[1] / (int64_t)normal / data[0]) : 0));
1110 /*ARGSUSED*/
1111 static int
1112 dt_print_stddev(dtrace_hdl_t *dtp, FILE *fp, caddr_t addr,
1113 size_t size, uint64_t normal)
1115 /* LINTED - alignment */
1116 uint64_t *data = (uint64_t *)addr;
1118 return (dt_printf(dtp, fp, " %16llu", data[0] ?
1119 (unsigned long long) dt_stddev(data, normal) : 0));
1122 /*ARGSUSED*/
1123 static int
1124 dt_print_bytes(dtrace_hdl_t *dtp, FILE *fp, caddr_t addr,
1125 size_t nbytes, int width, int quiet, int forceraw)
1128 * If the byte stream is a series of printable characters, followed by
1129 * a terminating byte, we print it out as a string. Otherwise, we
1130 * assume that it's something else and just print the bytes.
1132 int i, j, margin = 5;
1133 char *c = (char *)addr;
1135 if (nbytes == 0)
1136 return (0);
1138 if (forceraw)
1139 goto raw;
1141 if (dtp->dt_options[DTRACEOPT_RAWBYTES] != DTRACEOPT_UNSET)
1142 goto raw;
1144 for (i = 0; i < nbytes; i++) {
1146 * We define a "printable character" to be one for which
1147 * isprint(3C) returns non-zero, isspace(3C) returns non-zero,
1148 * or a character which is either backspace or the bell.
1149 * Backspace and the bell are regrettably special because
1150 * they fail the first two tests -- and yet they are entirely
1151 * printable. These are the only two control characters that
1152 * have meaning for the terminal and for which isprint(3C) and
1153 * isspace(3C) return 0.
1155 if (isprint(c[i]) || isspace(c[i]) ||
1156 c[i] == '\b' || c[i] == '\a')
1157 continue;
1159 if (c[i] == '\0' && i > 0) {
1161 * This looks like it might be a string. Before we
1162 * assume that it is indeed a string, check the
1163 * remainder of the byte range; if it contains
1164 * additional non-nul characters, we'll assume that
1165 * it's a binary stream that just happens to look like
1166 * a string, and we'll print out the individual bytes.
1168 for (j = i + 1; j < nbytes; j++) {
1169 if (c[j] != '\0')
1170 break;
1173 if (j != nbytes)
1174 break;
1176 if (quiet) {
1177 return (dt_printf(dtp, fp, "%s", c));
1178 } else {
1179 return (dt_printf(dtp, fp, " %s%*s",
1180 width < 0 ? " " : "", width, c));
1184 break;
1187 if (i == nbytes) {
1189 * The byte range is all printable characters, but there is
1190 * no trailing nul byte. We'll assume that it's a string and
1191 * print it as such.
1193 char *s = alloca(nbytes + 1);
1194 bcopy(c, s, nbytes);
1195 s[nbytes] = '\0';
1196 return (dt_printf(dtp, fp, " %-*s", width, s));
1199 raw:
1200 if (dt_printf(dtp, fp, "\n%*s ", margin, "") < 0)
1201 return (-1);
1203 for (i = 0; i < 16; i++)
1204 if (dt_printf(dtp, fp, " %c", "0123456789abcdef"[i]) < 0)
1205 return (-1);
1207 if (dt_printf(dtp, fp, " 0123456789abcdef\n") < 0)
1208 return (-1);
1211 for (i = 0; i < nbytes; i += 16) {
1212 if (dt_printf(dtp, fp, "%*s%5x:", margin, "", i) < 0)
1213 return (-1);
1215 for (j = i; j < i + 16 && j < nbytes; j++) {
1216 if (dt_printf(dtp, fp, " %02x", (uchar_t)c[j]) < 0)
1217 return (-1);
1220 while (j++ % 16) {
1221 if (dt_printf(dtp, fp, " ") < 0)
1222 return (-1);
1225 if (dt_printf(dtp, fp, " ") < 0)
1226 return (-1);
1228 for (j = i; j < i + 16 && j < nbytes; j++) {
1229 if (dt_printf(dtp, fp, "%c",
1230 c[j] < ' ' || c[j] > '~' ? '.' : c[j]) < 0)
1231 return (-1);
1234 if (dt_printf(dtp, fp, "\n") < 0)
1235 return (-1);
1238 return (0);
1242 dt_print_stack(dtrace_hdl_t *dtp, FILE *fp, const char *format,
1243 caddr_t addr, int depth, int size)
1245 dtrace_syminfo_t dts;
1246 GElf_Sym sym;
1247 int i, indent;
1248 char c[PATH_MAX * 2];
1249 uint64_t pc;
1251 if (dt_printf(dtp, fp, "\n") < 0)
1252 return (-1);
1254 if (format == NULL)
1255 format = "%s";
1257 if (dtp->dt_options[DTRACEOPT_STACKINDENT] != DTRACEOPT_UNSET)
1258 indent = (int)dtp->dt_options[DTRACEOPT_STACKINDENT];
1259 else
1260 indent = _dtrace_stkindent;
1262 for (i = 0; i < depth; i++) {
1263 switch (size) {
1264 case sizeof (uint32_t):
1265 /* LINTED - alignment */
1266 pc = *((uint32_t *)addr);
1267 break;
1269 case sizeof (uint64_t):
1270 /* LINTED - alignment */
1271 pc = *((uint64_t *)addr);
1272 break;
1274 default:
1275 return (dt_set_errno(dtp, EDT_BADSTACKPC));
1278 if (pc == NULL)
1279 break;
1281 addr += size;
1283 if (dt_printf(dtp, fp, "%*s", indent, "") < 0)
1284 return (-1);
1286 if (dtrace_lookup_by_addr(dtp, pc, &sym, &dts) == 0) {
1287 if (pc > sym.st_value) {
1288 (void) snprintf(c, sizeof (c), "%s`%s+0x%llx",
1289 dts.dts_object, dts.dts_name,
1290 pc - sym.st_value);
1291 } else {
1292 (void) snprintf(c, sizeof (c), "%s`%s",
1293 dts.dts_object, dts.dts_name);
1295 } else {
1297 * We'll repeat the lookup, but this time we'll specify
1298 * a NULL GElf_Sym -- indicating that we're only
1299 * interested in the containing module.
1301 if (dtrace_lookup_by_addr(dtp, pc, NULL, &dts) == 0) {
1302 (void) snprintf(c, sizeof (c), "%s`0x%llx",
1303 dts.dts_object, pc);
1304 } else {
1305 (void) snprintf(c, sizeof (c), "0x%llx", pc);
1309 if (dt_printf(dtp, fp, format, c) < 0)
1310 return (-1);
1312 if (dt_printf(dtp, fp, "\n") < 0)
1313 return (-1);
1316 return (0);
1320 dt_print_ustack(dtrace_hdl_t *dtp, FILE *fp, const char *format,
1321 caddr_t addr, uint64_t arg)
1323 /* LINTED - alignment */
1324 uint64_t *pc = (uint64_t *)addr;
1325 uint32_t depth = DTRACE_USTACK_NFRAMES(arg);
1326 uint32_t strsize = DTRACE_USTACK_STRSIZE(arg);
1327 const char *strbase = addr + (depth + 1) * sizeof (uint64_t);
1328 const char *str = strsize ? strbase : NULL;
1329 int err = 0;
1331 char name[PATH_MAX], objname[PATH_MAX], c[PATH_MAX * 2];
1332 struct ps_prochandle *P;
1333 GElf_Sym sym;
1334 int i, indent;
1335 pid_t pid;
1337 if (depth == 0)
1338 return (0);
1340 pid = (pid_t)*pc++;
1342 if (dt_printf(dtp, fp, "\n") < 0)
1343 return (-1);
1345 if (format == NULL)
1346 format = "%s";
1348 if (dtp->dt_options[DTRACEOPT_STACKINDENT] != DTRACEOPT_UNSET)
1349 indent = (int)dtp->dt_options[DTRACEOPT_STACKINDENT];
1350 else
1351 indent = _dtrace_stkindent;
1354 * Ultimately, we need to add an entry point in the library vector for
1355 * determining <symbol, offset> from <pid, address>. For now, if
1356 * this is a vector open, we just print the raw address or string.
1358 if (dtp->dt_vector == NULL)
1359 P = dt_proc_grab(dtp, pid, PGRAB_RDONLY | PGRAB_FORCE, 0);
1360 else
1361 P = NULL;
1363 if (P != NULL)
1364 dt_proc_lock(dtp, P); /* lock handle while we perform lookups */
1366 for (i = 0; i < depth && pc[i] != NULL; i++) {
1367 const prmap_t *map;
1369 if ((err = dt_printf(dtp, fp, "%*s", indent, "")) < 0)
1370 break;
1372 if (P != NULL && Plookup_by_addr(P, pc[i],
1373 name, sizeof (name), &sym) == 0) {
1374 (void) Pobjname(P, pc[i], objname, sizeof (objname));
1376 if (pc[i] > sym.st_value) {
1377 (void) snprintf(c, sizeof (c),
1378 "%s`%s+0x%llx", dt_basename(objname), name,
1379 (u_longlong_t)(pc[i] - sym.st_value));
1380 } else {
1381 (void) snprintf(c, sizeof (c),
1382 "%s`%s", dt_basename(objname), name);
1384 } else if (str != NULL && str[0] != '\0' && str[0] != '@' &&
1385 (P != NULL && ((map = Paddr_to_map(P, pc[i])) == NULL ||
1386 (map->pr_mflags & MA_WRITE)))) {
1388 * If the current string pointer in the string table
1389 * does not point to an empty string _and_ the program
1390 * counter falls in a writable region, we'll use the
1391 * string from the string table instead of the raw
1392 * address. This last condition is necessary because
1393 * some (broken) ustack helpers will return a string
1394 * even for a program counter that they can't
1395 * identify. If we have a string for a program
1396 * counter that falls in a segment that isn't
1397 * writable, we assume that we have fallen into this
1398 * case and we refuse to use the string.
1400 (void) snprintf(c, sizeof (c), "%s", str);
1401 } else {
1402 if (P != NULL && Pobjname(P, pc[i], objname,
1403 sizeof (objname)) != NULL) {
1404 (void) snprintf(c, sizeof (c), "%s`0x%llx",
1405 dt_basename(objname), (u_longlong_t)pc[i]);
1406 } else {
1407 (void) snprintf(c, sizeof (c), "0x%llx",
1408 (u_longlong_t)pc[i]);
1412 if ((err = dt_printf(dtp, fp, format, c)) < 0)
1413 break;
1415 if ((err = dt_printf(dtp, fp, "\n")) < 0)
1416 break;
1418 if (str != NULL && str[0] == '@') {
1420 * If the first character of the string is an "at" sign,
1421 * then the string is inferred to be an annotation --
1422 * and it is printed out beneath the frame and offset
1423 * with brackets.
1425 if ((err = dt_printf(dtp, fp, "%*s", indent, "")) < 0)
1426 break;
1428 (void) snprintf(c, sizeof (c), " [ %s ]", &str[1]);
1430 if ((err = dt_printf(dtp, fp, format, c)) < 0)
1431 break;
1433 if ((err = dt_printf(dtp, fp, "\n")) < 0)
1434 break;
1437 if (str != NULL) {
1438 str += strlen(str) + 1;
1439 if (str - strbase >= strsize)
1440 str = NULL;
1444 if (P != NULL) {
1445 dt_proc_unlock(dtp, P);
1446 dt_proc_release(dtp, P);
1449 return (err);
1452 static int
1453 dt_print_usym(dtrace_hdl_t *dtp, FILE *fp, caddr_t addr, dtrace_actkind_t act)
1455 /* LINTED - alignment */
1456 uint64_t pid = ((uint64_t *)addr)[0];
1457 /* LINTED - alignment */
1458 uint64_t pc = ((uint64_t *)addr)[1];
1459 const char *format = " %-50s";
1460 char *s;
1461 int n, len = 256;
1463 if (act == DTRACEACT_USYM && dtp->dt_vector == NULL) {
1464 struct ps_prochandle *P;
1466 if ((P = dt_proc_grab(dtp, pid,
1467 PGRAB_RDONLY | PGRAB_FORCE, 0)) != NULL) {
1468 GElf_Sym sym;
1470 dt_proc_lock(dtp, P);
1472 if (Plookup_by_addr(P, pc, NULL, 0, &sym) == 0)
1473 pc = sym.st_value;
1475 dt_proc_unlock(dtp, P);
1476 dt_proc_release(dtp, P);
1480 do {
1481 n = len;
1482 s = alloca(n);
1483 } while ((len = dtrace_uaddr2str(dtp, pid, pc, s, n)) > n);
1485 return (dt_printf(dtp, fp, format, s));
1489 dt_print_umod(dtrace_hdl_t *dtp, FILE *fp, const char *format, caddr_t addr)
1491 /* LINTED - alignment */
1492 uint64_t pid = ((uint64_t *)addr)[0];
1493 /* LINTED - alignment */
1494 uint64_t pc = ((uint64_t *)addr)[1];
1495 int err = 0;
1497 char objname[PATH_MAX], c[PATH_MAX * 2];
1498 struct ps_prochandle *P;
1500 if (format == NULL)
1501 format = " %-50s";
1504 * See the comment in dt_print_ustack() for the rationale for
1505 * printing raw addresses in the vectored case.
1507 if (dtp->dt_vector == NULL)
1508 P = dt_proc_grab(dtp, pid, PGRAB_RDONLY | PGRAB_FORCE, 0);
1509 else
1510 P = NULL;
1512 if (P != NULL)
1513 dt_proc_lock(dtp, P); /* lock handle while we perform lookups */
1515 if (P != NULL && Pobjname(P, pc, objname, sizeof (objname)) != NULL) {
1516 (void) snprintf(c, sizeof (c), "%s", dt_basename(objname));
1517 } else {
1518 (void) snprintf(c, sizeof (c), "0x%llx", (u_longlong_t)pc);
1521 err = dt_printf(dtp, fp, format, c);
1523 if (P != NULL) {
1524 dt_proc_unlock(dtp, P);
1525 dt_proc_release(dtp, P);
1528 return (err);
1531 static int
1532 dt_print_sym(dtrace_hdl_t *dtp, FILE *fp, const char *format, caddr_t addr)
1534 /* LINTED - alignment */
1535 uint64_t pc = *((uint64_t *)addr);
1536 dtrace_syminfo_t dts;
1537 GElf_Sym sym;
1538 char c[PATH_MAX * 2];
1540 if (format == NULL)
1541 format = " %-50s";
1543 if (dtrace_lookup_by_addr(dtp, pc, &sym, &dts) == 0) {
1544 (void) snprintf(c, sizeof (c), "%s`%s",
1545 dts.dts_object, dts.dts_name);
1546 } else {
1548 * We'll repeat the lookup, but this time we'll specify a
1549 * NULL GElf_Sym -- indicating that we're only interested in
1550 * the containing module.
1552 if (dtrace_lookup_by_addr(dtp, pc, NULL, &dts) == 0) {
1553 (void) snprintf(c, sizeof (c), "%s`0x%llx",
1554 dts.dts_object, (u_longlong_t)pc);
1555 } else {
1556 (void) snprintf(c, sizeof (c), "0x%llx",
1557 (u_longlong_t)pc);
1561 if (dt_printf(dtp, fp, format, c) < 0)
1562 return (-1);
1564 return (0);
1568 dt_print_mod(dtrace_hdl_t *dtp, FILE *fp, const char *format, caddr_t addr)
1570 /* LINTED - alignment */
1571 uint64_t pc = *((uint64_t *)addr);
1572 dtrace_syminfo_t dts;
1573 char c[PATH_MAX * 2];
1575 if (format == NULL)
1576 format = " %-50s";
1578 if (dtrace_lookup_by_addr(dtp, pc, NULL, &dts) == 0) {
1579 (void) snprintf(c, sizeof (c), "%s", dts.dts_object);
1580 } else {
1581 (void) snprintf(c, sizeof (c), "0x%llx", (u_longlong_t)pc);
1584 if (dt_printf(dtp, fp, format, c) < 0)
1585 return (-1);
1587 return (0);
1590 typedef struct dt_normal {
1591 dtrace_aggvarid_t dtnd_id;
1592 uint64_t dtnd_normal;
1593 } dt_normal_t;
1595 static int
1596 dt_normalize_agg(const dtrace_aggdata_t *aggdata, void *arg)
1598 dt_normal_t *normal = arg;
1599 dtrace_aggdesc_t *agg = aggdata->dtada_desc;
1600 dtrace_aggvarid_t id = normal->dtnd_id;
1602 if (agg->dtagd_nrecs == 0)
1603 return (DTRACE_AGGWALK_NEXT);
1605 if (agg->dtagd_varid != id)
1606 return (DTRACE_AGGWALK_NEXT);
1608 ((dtrace_aggdata_t *)aggdata)->dtada_normal = normal->dtnd_normal;
1609 return (DTRACE_AGGWALK_NORMALIZE);
1612 static int
1613 dt_normalize(dtrace_hdl_t *dtp, caddr_t base, dtrace_recdesc_t *rec)
1615 dt_normal_t normal;
1616 caddr_t addr;
1619 * We (should) have two records: the aggregation ID followed by the
1620 * normalization value.
1622 addr = base + rec->dtrd_offset;
1624 if (rec->dtrd_size != sizeof (dtrace_aggvarid_t))
1625 return (dt_set_errno(dtp, EDT_BADNORMAL));
1627 /* LINTED - alignment */
1628 normal.dtnd_id = *((dtrace_aggvarid_t *)addr);
1629 rec++;
1631 if (rec->dtrd_action != DTRACEACT_LIBACT)
1632 return (dt_set_errno(dtp, EDT_BADNORMAL));
1634 if (rec->dtrd_arg != DT_ACT_NORMALIZE)
1635 return (dt_set_errno(dtp, EDT_BADNORMAL));
1637 addr = base + rec->dtrd_offset;
1639 switch (rec->dtrd_size) {
1640 case sizeof (uint64_t):
1641 /* LINTED - alignment */
1642 normal.dtnd_normal = *((uint64_t *)addr);
1643 break;
1644 case sizeof (uint32_t):
1645 /* LINTED - alignment */
1646 normal.dtnd_normal = *((uint32_t *)addr);
1647 break;
1648 case sizeof (uint16_t):
1649 /* LINTED - alignment */
1650 normal.dtnd_normal = *((uint16_t *)addr);
1651 break;
1652 case sizeof (uint8_t):
1653 normal.dtnd_normal = *((uint8_t *)addr);
1654 break;
1655 default:
1656 return (dt_set_errno(dtp, EDT_BADNORMAL));
1659 (void) dtrace_aggregate_walk(dtp, dt_normalize_agg, &normal);
1661 return (0);
1664 static int
1665 dt_denormalize_agg(const dtrace_aggdata_t *aggdata, void *arg)
1667 dtrace_aggdesc_t *agg = aggdata->dtada_desc;
1668 dtrace_aggvarid_t id = *((dtrace_aggvarid_t *)arg);
1670 if (agg->dtagd_nrecs == 0)
1671 return (DTRACE_AGGWALK_NEXT);
1673 if (agg->dtagd_varid != id)
1674 return (DTRACE_AGGWALK_NEXT);
1676 return (DTRACE_AGGWALK_DENORMALIZE);
1679 static int
1680 dt_clear_agg(const dtrace_aggdata_t *aggdata, void *arg)
1682 dtrace_aggdesc_t *agg = aggdata->dtada_desc;
1683 dtrace_aggvarid_t id = *((dtrace_aggvarid_t *)arg);
1685 if (agg->dtagd_nrecs == 0)
1686 return (DTRACE_AGGWALK_NEXT);
1688 if (agg->dtagd_varid != id)
1689 return (DTRACE_AGGWALK_NEXT);
1691 return (DTRACE_AGGWALK_CLEAR);
1694 typedef struct dt_trunc {
1695 dtrace_aggvarid_t dttd_id;
1696 uint64_t dttd_remaining;
1697 } dt_trunc_t;
1699 static int
1700 dt_trunc_agg(const dtrace_aggdata_t *aggdata, void *arg)
1702 dt_trunc_t *trunc = arg;
1703 dtrace_aggdesc_t *agg = aggdata->dtada_desc;
1704 dtrace_aggvarid_t id = trunc->dttd_id;
1706 if (agg->dtagd_nrecs == 0)
1707 return (DTRACE_AGGWALK_NEXT);
1709 if (agg->dtagd_varid != id)
1710 return (DTRACE_AGGWALK_NEXT);
1712 if (trunc->dttd_remaining == 0)
1713 return (DTRACE_AGGWALK_REMOVE);
1715 trunc->dttd_remaining--;
1716 return (DTRACE_AGGWALK_NEXT);
1719 static int
1720 dt_trunc(dtrace_hdl_t *dtp, caddr_t base, dtrace_recdesc_t *rec)
1722 dt_trunc_t trunc;
1723 caddr_t addr;
1724 int64_t remaining;
1725 int (*func)(dtrace_hdl_t *, dtrace_aggregate_f *, void *);
1728 * We (should) have two records: the aggregation ID followed by the
1729 * number of aggregation entries after which the aggregation is to be
1730 * truncated.
1732 addr = base + rec->dtrd_offset;
1734 if (rec->dtrd_size != sizeof (dtrace_aggvarid_t))
1735 return (dt_set_errno(dtp, EDT_BADTRUNC));
1737 /* LINTED - alignment */
1738 trunc.dttd_id = *((dtrace_aggvarid_t *)addr);
1739 rec++;
1741 if (rec->dtrd_action != DTRACEACT_LIBACT)
1742 return (dt_set_errno(dtp, EDT_BADTRUNC));
1744 if (rec->dtrd_arg != DT_ACT_TRUNC)
1745 return (dt_set_errno(dtp, EDT_BADTRUNC));
1747 addr = base + rec->dtrd_offset;
1749 switch (rec->dtrd_size) {
1750 case sizeof (uint64_t):
1751 /* LINTED - alignment */
1752 remaining = *((int64_t *)addr);
1753 break;
1754 case sizeof (uint32_t):
1755 /* LINTED - alignment */
1756 remaining = *((int32_t *)addr);
1757 break;
1758 case sizeof (uint16_t):
1759 /* LINTED - alignment */
1760 remaining = *((int16_t *)addr);
1761 break;
1762 case sizeof (uint8_t):
1763 remaining = *((int8_t *)addr);
1764 break;
1765 default:
1766 return (dt_set_errno(dtp, EDT_BADNORMAL));
1769 if (remaining < 0) {
1770 func = dtrace_aggregate_walk_valsorted;
1771 remaining = -remaining;
1772 } else {
1773 func = dtrace_aggregate_walk_valrevsorted;
1776 assert(remaining >= 0);
1777 trunc.dttd_remaining = remaining;
1779 (void) func(dtp, dt_trunc_agg, &trunc);
1781 return (0);
1784 static int
1785 dt_print_datum(dtrace_hdl_t *dtp, FILE *fp, dtrace_recdesc_t *rec,
1786 caddr_t addr, size_t size, const dtrace_aggdata_t *aggdata,
1787 uint64_t normal, dt_print_aggdata_t *pd)
1789 int err, width;
1790 dtrace_actkind_t act = rec->dtrd_action;
1791 boolean_t packed = pd->dtpa_agghist || pd->dtpa_aggpack;
1792 dtrace_aggdesc_t *agg = aggdata->dtada_desc;
1794 static struct {
1795 size_t size;
1796 int width;
1797 int packedwidth;
1798 } *fmt, fmttab[] = {
1799 { sizeof (uint8_t), 3, 3 },
1800 { sizeof (uint16_t), 5, 5 },
1801 { sizeof (uint32_t), 8, 8 },
1802 { sizeof (uint64_t), 16, 16 },
1803 { 0, -50, 16 }
1806 if (packed && pd->dtpa_agghisthdr != agg->dtagd_varid) {
1807 dtrace_recdesc_t *r;
1809 width = 0;
1812 * To print our quantization header for either an agghist or
1813 * aggpack aggregation, we need to iterate through all of our
1814 * of our records to determine their width.
1816 for (r = rec; !DTRACEACT_ISAGG(r->dtrd_action); r++) {
1817 for (fmt = fmttab; fmt->size &&
1818 fmt->size != r->dtrd_size; fmt++)
1819 continue;
1821 width += fmt->packedwidth + 1;
1824 if (pd->dtpa_agghist) {
1825 if (dt_print_quanthdr(dtp, fp, width) < 0)
1826 return (-1);
1827 } else {
1828 if (dt_print_quanthdr_packed(dtp, fp,
1829 width, aggdata, r->dtrd_action) < 0)
1830 return (-1);
1833 pd->dtpa_agghisthdr = agg->dtagd_varid;
1836 if (pd->dtpa_agghist && DTRACEACT_ISAGG(act)) {
1837 char positives = aggdata->dtada_flags & DTRACE_A_HASPOSITIVES;
1838 char negatives = aggdata->dtada_flags & DTRACE_A_HASNEGATIVES;
1839 int64_t val;
1841 assert(act == DTRACEAGG_SUM || act == DTRACEAGG_COUNT);
1842 val = (long long)*((uint64_t *)addr);
1844 if (dt_printf(dtp, fp, " ") < 0)
1845 return (-1);
1847 return (dt_print_quantline(dtp, fp, val, normal,
1848 aggdata->dtada_total, positives, negatives));
1851 if (pd->dtpa_aggpack && DTRACEACT_ISAGG(act)) {
1852 switch (act) {
1853 case DTRACEAGG_QUANTIZE:
1854 return (dt_print_quantize_packed(dtp,
1855 fp, addr, size, aggdata));
1856 case DTRACEAGG_LQUANTIZE:
1857 return (dt_print_lquantize_packed(dtp,
1858 fp, addr, size, aggdata));
1859 default:
1860 break;
1864 switch (act) {
1865 case DTRACEACT_STACK:
1866 return (dt_print_stack(dtp, fp, NULL, addr,
1867 rec->dtrd_arg, rec->dtrd_size / rec->dtrd_arg));
1869 case DTRACEACT_USTACK:
1870 case DTRACEACT_JSTACK:
1871 return (dt_print_ustack(dtp, fp, NULL, addr, rec->dtrd_arg));
1873 case DTRACEACT_USYM:
1874 case DTRACEACT_UADDR:
1875 return (dt_print_usym(dtp, fp, addr, act));
1877 case DTRACEACT_UMOD:
1878 return (dt_print_umod(dtp, fp, NULL, addr));
1880 case DTRACEACT_SYM:
1881 return (dt_print_sym(dtp, fp, NULL, addr));
1883 case DTRACEACT_MOD:
1884 return (dt_print_mod(dtp, fp, NULL, addr));
1886 case DTRACEAGG_QUANTIZE:
1887 return (dt_print_quantize(dtp, fp, addr, size, normal));
1889 case DTRACEAGG_LQUANTIZE:
1890 return (dt_print_lquantize(dtp, fp, addr, size, normal));
1892 case DTRACEAGG_LLQUANTIZE:
1893 return (dt_print_llquantize(dtp, fp, addr, size, normal));
1895 case DTRACEAGG_AVG:
1896 return (dt_print_average(dtp, fp, addr, size, normal));
1898 case DTRACEAGG_STDDEV:
1899 return (dt_print_stddev(dtp, fp, addr, size, normal));
1901 default:
1902 break;
1905 for (fmt = fmttab; fmt->size && fmt->size != size; fmt++)
1906 continue;
1908 width = packed ? fmt->packedwidth : fmt->width;
1910 switch (size) {
1911 case sizeof (uint64_t):
1912 err = dt_printf(dtp, fp, " %*lld", width,
1913 /* LINTED - alignment */
1914 (long long)*((uint64_t *)addr) / normal);
1915 break;
1916 case sizeof (uint32_t):
1917 /* LINTED - alignment */
1918 err = dt_printf(dtp, fp, " %*d", width, *((uint32_t *)addr) /
1919 (uint32_t)normal);
1920 break;
1921 case sizeof (uint16_t):
1922 /* LINTED - alignment */
1923 err = dt_printf(dtp, fp, " %*d", width, *((uint16_t *)addr) /
1924 (uint32_t)normal);
1925 break;
1926 case sizeof (uint8_t):
1927 err = dt_printf(dtp, fp, " %*d", width, *((uint8_t *)addr) /
1928 (uint32_t)normal);
1929 break;
1930 default:
1931 err = dt_print_bytes(dtp, fp, addr, size, width, 0, 0);
1932 break;
1935 return (err);
1939 dt_print_aggs(const dtrace_aggdata_t **aggsdata, int naggvars, void *arg)
1941 int i, aggact = 0;
1942 dt_print_aggdata_t *pd = arg;
1943 const dtrace_aggdata_t *aggdata = aggsdata[0];
1944 dtrace_aggdesc_t *agg = aggdata->dtada_desc;
1945 FILE *fp = pd->dtpa_fp;
1946 dtrace_hdl_t *dtp = pd->dtpa_dtp;
1947 dtrace_recdesc_t *rec;
1948 dtrace_actkind_t act;
1949 caddr_t addr;
1950 size_t size;
1952 pd->dtpa_agghist = (aggdata->dtada_flags & DTRACE_A_TOTAL);
1953 pd->dtpa_aggpack = (aggdata->dtada_flags & DTRACE_A_MINMAXBIN);
1956 * Iterate over each record description in the key, printing the traced
1957 * data, skipping the first datum (the tuple member created by the
1958 * compiler).
1960 for (i = 1; i < agg->dtagd_nrecs; i++) {
1961 rec = &agg->dtagd_rec[i];
1962 act = rec->dtrd_action;
1963 addr = aggdata->dtada_data + rec->dtrd_offset;
1964 size = rec->dtrd_size;
1966 if (DTRACEACT_ISAGG(act)) {
1967 aggact = i;
1968 break;
1971 if (dt_print_datum(dtp, fp, rec, addr,
1972 size, aggdata, 1, pd) < 0)
1973 return (-1);
1975 if (dt_buffered_flush(dtp, NULL, rec, aggdata,
1976 DTRACE_BUFDATA_AGGKEY) < 0)
1977 return (-1);
1980 assert(aggact != 0);
1982 for (i = (naggvars == 1 ? 0 : 1); i < naggvars; i++) {
1983 uint64_t normal;
1985 aggdata = aggsdata[i];
1986 agg = aggdata->dtada_desc;
1987 rec = &agg->dtagd_rec[aggact];
1988 act = rec->dtrd_action;
1989 addr = aggdata->dtada_data + rec->dtrd_offset;
1990 size = rec->dtrd_size;
1992 assert(DTRACEACT_ISAGG(act));
1993 normal = aggdata->dtada_normal;
1995 if (dt_print_datum(dtp, fp, rec, addr,
1996 size, aggdata, normal, pd) < 0)
1997 return (-1);
1999 if (dt_buffered_flush(dtp, NULL, rec, aggdata,
2000 DTRACE_BUFDATA_AGGVAL) < 0)
2001 return (-1);
2003 if (!pd->dtpa_allunprint)
2004 agg->dtagd_flags |= DTRACE_AGD_PRINTED;
2007 if (!pd->dtpa_agghist && !pd->dtpa_aggpack) {
2008 if (dt_printf(dtp, fp, "\n") < 0)
2009 return (-1);
2012 if (dt_buffered_flush(dtp, NULL, NULL, aggdata,
2013 DTRACE_BUFDATA_AGGFORMAT | DTRACE_BUFDATA_AGGLAST) < 0)
2014 return (-1);
2016 return (0);
2020 dt_print_agg(const dtrace_aggdata_t *aggdata, void *arg)
2022 dt_print_aggdata_t *pd = arg;
2023 dtrace_aggdesc_t *agg = aggdata->dtada_desc;
2024 dtrace_aggvarid_t aggvarid = pd->dtpa_id;
2026 if (pd->dtpa_allunprint) {
2027 if (agg->dtagd_flags & DTRACE_AGD_PRINTED)
2028 return (0);
2029 } else {
2031 * If we're not printing all unprinted aggregations, then the
2032 * aggregation variable ID denotes a specific aggregation
2033 * variable that we should print -- skip any other aggregations
2034 * that we encounter.
2036 if (agg->dtagd_nrecs == 0)
2037 return (0);
2039 if (aggvarid != agg->dtagd_varid)
2040 return (0);
2043 return (dt_print_aggs(&aggdata, 1, arg));
2047 dt_setopt(dtrace_hdl_t *dtp, const dtrace_probedata_t *data,
2048 const char *option, const char *value)
2050 int len, rval;
2051 char *msg;
2052 const char *errstr;
2053 dtrace_setoptdata_t optdata;
2055 bzero(&optdata, sizeof (optdata));
2056 (void) dtrace_getopt(dtp, option, &optdata.dtsda_oldval);
2058 if (dtrace_setopt(dtp, option, value) == 0) {
2059 (void) dtrace_getopt(dtp, option, &optdata.dtsda_newval);
2060 optdata.dtsda_probe = data;
2061 optdata.dtsda_option = option;
2062 optdata.dtsda_handle = dtp;
2064 if ((rval = dt_handle_setopt(dtp, &optdata)) != 0)
2065 return (rval);
2067 return (0);
2070 errstr = dtrace_errmsg(dtp, dtrace_errno(dtp));
2071 len = strlen(option) + strlen(value) + strlen(errstr) + 80;
2072 msg = alloca(len);
2074 (void) snprintf(msg, len, "couldn't set option \"%s\" to \"%s\": %s\n",
2075 option, value, errstr);
2077 if ((rval = dt_handle_liberr(dtp, data, msg)) == 0)
2078 return (0);
2080 return (rval);
2083 static int
2084 dt_consume_cpu(dtrace_hdl_t *dtp, FILE *fp, int cpu,
2085 dtrace_bufdesc_t *buf, boolean_t just_one,
2086 dtrace_consume_probe_f *efunc, dtrace_consume_rec_f *rfunc, void *arg)
2088 dtrace_epid_t id;
2089 size_t offs;
2090 int flow = (dtp->dt_options[DTRACEOPT_FLOWINDENT] != DTRACEOPT_UNSET);
2091 int quiet = (dtp->dt_options[DTRACEOPT_QUIET] != DTRACEOPT_UNSET);
2092 int rval, i, n;
2093 uint64_t tracememsize = 0;
2094 dtrace_probedata_t data;
2095 uint64_t drops;
2097 bzero(&data, sizeof (data));
2098 data.dtpda_handle = dtp;
2099 data.dtpda_cpu = cpu;
2100 data.dtpda_flow = dtp->dt_flow;
2101 data.dtpda_indent = dtp->dt_indent;
2102 data.dtpda_prefix = dtp->dt_prefix;
2104 for (offs = buf->dtbd_oldest; offs < buf->dtbd_size; ) {
2105 dtrace_eprobedesc_t *epd;
2108 * We're guaranteed to have an ID.
2110 id = *(uint32_t *)((uintptr_t)buf->dtbd_data + offs);
2112 if (id == DTRACE_EPIDNONE) {
2114 * This is filler to assure proper alignment of the
2115 * next record; we simply ignore it.
2117 offs += sizeof (id);
2118 continue;
2121 if ((rval = dt_epid_lookup(dtp, id, &data.dtpda_edesc,
2122 &data.dtpda_pdesc)) != 0)
2123 return (rval);
2125 epd = data.dtpda_edesc;
2126 data.dtpda_data = buf->dtbd_data + offs;
2128 if (data.dtpda_edesc->dtepd_uarg != DT_ECB_DEFAULT) {
2129 rval = dt_handle(dtp, &data);
2131 if (rval == DTRACE_CONSUME_NEXT)
2132 goto nextepid;
2134 if (rval == DTRACE_CONSUME_ERROR)
2135 return (-1);
2138 if (flow)
2139 (void) dt_flowindent(dtp, &data, dtp->dt_last_epid,
2140 buf, offs);
2142 rval = (*efunc)(&data, arg);
2144 if (flow) {
2145 if (data.dtpda_flow == DTRACEFLOW_ENTRY)
2146 data.dtpda_indent += 2;
2149 if (rval == DTRACE_CONSUME_NEXT)
2150 goto nextepid;
2152 if (rval == DTRACE_CONSUME_ABORT)
2153 return (dt_set_errno(dtp, EDT_DIRABORT));
2155 if (rval != DTRACE_CONSUME_THIS)
2156 return (dt_set_errno(dtp, EDT_BADRVAL));
2158 for (i = 0; i < epd->dtepd_nrecs; i++) {
2159 caddr_t addr;
2160 dtrace_recdesc_t *rec = &epd->dtepd_rec[i];
2161 dtrace_actkind_t act = rec->dtrd_action;
2163 data.dtpda_data = buf->dtbd_data + offs +
2164 rec->dtrd_offset;
2165 addr = data.dtpda_data;
2167 if (act == DTRACEACT_LIBACT) {
2168 uint64_t arg = rec->dtrd_arg;
2169 dtrace_aggvarid_t id;
2171 switch (arg) {
2172 case DT_ACT_CLEAR:
2173 /* LINTED - alignment */
2174 id = *((dtrace_aggvarid_t *)addr);
2175 (void) dtrace_aggregate_walk(dtp,
2176 dt_clear_agg, &id);
2177 continue;
2179 case DT_ACT_DENORMALIZE:
2180 /* LINTED - alignment */
2181 id = *((dtrace_aggvarid_t *)addr);
2182 (void) dtrace_aggregate_walk(dtp,
2183 dt_denormalize_agg, &id);
2184 continue;
2186 case DT_ACT_FTRUNCATE:
2187 if (fp == NULL)
2188 continue;
2190 (void) fflush(fp);
2191 (void) ftruncate(fileno(fp), 0);
2192 (void) fseeko(fp, 0, SEEK_SET);
2193 continue;
2195 case DT_ACT_NORMALIZE:
2196 if (i == epd->dtepd_nrecs - 1)
2197 return (dt_set_errno(dtp,
2198 EDT_BADNORMAL));
2200 if (dt_normalize(dtp,
2201 buf->dtbd_data + offs, rec) != 0)
2202 return (-1);
2204 i++;
2205 continue;
2207 case DT_ACT_SETOPT: {
2208 uint64_t *opts = dtp->dt_options;
2209 dtrace_recdesc_t *valrec;
2210 uint32_t valsize;
2211 caddr_t val;
2212 int rv;
2214 if (i == epd->dtepd_nrecs - 1) {
2215 return (dt_set_errno(dtp,
2216 EDT_BADSETOPT));
2219 valrec = &epd->dtepd_rec[++i];
2220 valsize = valrec->dtrd_size;
2222 if (valrec->dtrd_action != act ||
2223 valrec->dtrd_arg != arg) {
2224 return (dt_set_errno(dtp,
2225 EDT_BADSETOPT));
2228 if (valsize > sizeof (uint64_t)) {
2229 val = buf->dtbd_data + offs +
2230 valrec->dtrd_offset;
2231 } else {
2232 val = "1";
2235 rv = dt_setopt(dtp, &data, addr, val);
2237 if (rv != 0)
2238 return (-1);
2240 flow = (opts[DTRACEOPT_FLOWINDENT] !=
2241 DTRACEOPT_UNSET);
2242 quiet = (opts[DTRACEOPT_QUIET] !=
2243 DTRACEOPT_UNSET);
2245 continue;
2248 case DT_ACT_TRUNC:
2249 if (i == epd->dtepd_nrecs - 1)
2250 return (dt_set_errno(dtp,
2251 EDT_BADTRUNC));
2253 if (dt_trunc(dtp,
2254 buf->dtbd_data + offs, rec) != 0)
2255 return (-1);
2257 i++;
2258 continue;
2260 default:
2261 continue;
2265 if (act == DTRACEACT_TRACEMEM_DYNSIZE &&
2266 rec->dtrd_size == sizeof (uint64_t)) {
2267 /* LINTED - alignment */
2268 tracememsize = *((unsigned long long *)addr);
2269 continue;
2272 rval = (*rfunc)(&data, rec, arg);
2274 if (rval == DTRACE_CONSUME_NEXT)
2275 continue;
2277 if (rval == DTRACE_CONSUME_ABORT)
2278 return (dt_set_errno(dtp, EDT_DIRABORT));
2280 if (rval != DTRACE_CONSUME_THIS)
2281 return (dt_set_errno(dtp, EDT_BADRVAL));
2283 if (act == DTRACEACT_STACK) {
2284 int depth = rec->dtrd_arg;
2286 if (dt_print_stack(dtp, fp, NULL, addr, depth,
2287 rec->dtrd_size / depth) < 0)
2288 return (-1);
2289 goto nextrec;
2292 if (act == DTRACEACT_USTACK ||
2293 act == DTRACEACT_JSTACK) {
2294 if (dt_print_ustack(dtp, fp, NULL,
2295 addr, rec->dtrd_arg) < 0)
2296 return (-1);
2297 goto nextrec;
2300 if (act == DTRACEACT_SYM) {
2301 if (dt_print_sym(dtp, fp, NULL, addr) < 0)
2302 return (-1);
2303 goto nextrec;
2306 if (act == DTRACEACT_MOD) {
2307 if (dt_print_mod(dtp, fp, NULL, addr) < 0)
2308 return (-1);
2309 goto nextrec;
2312 if (act == DTRACEACT_USYM || act == DTRACEACT_UADDR) {
2313 if (dt_print_usym(dtp, fp, addr, act) < 0)
2314 return (-1);
2315 goto nextrec;
2318 if (act == DTRACEACT_UMOD) {
2319 if (dt_print_umod(dtp, fp, NULL, addr) < 0)
2320 return (-1);
2321 goto nextrec;
2324 if (DTRACEACT_ISPRINTFLIKE(act)) {
2325 void *fmtdata;
2326 int (*func)(dtrace_hdl_t *, FILE *, void *,
2327 const dtrace_probedata_t *,
2328 const dtrace_recdesc_t *, uint_t,
2329 const void *buf, size_t);
2331 if ((fmtdata = dt_format_lookup(dtp,
2332 rec->dtrd_format)) == NULL)
2333 goto nofmt;
2335 switch (act) {
2336 case DTRACEACT_PRINTF:
2337 func = dtrace_fprintf;
2338 break;
2339 case DTRACEACT_PRINTA:
2340 func = dtrace_fprinta;
2341 break;
2342 case DTRACEACT_SYSTEM:
2343 func = dtrace_system;
2344 break;
2345 case DTRACEACT_FREOPEN:
2346 func = dtrace_freopen;
2347 break;
2350 n = (*func)(dtp, fp, fmtdata, &data,
2351 rec, epd->dtepd_nrecs - i,
2352 (uchar_t *)buf->dtbd_data + offs,
2353 buf->dtbd_size - offs);
2355 if (n < 0)
2356 return (-1); /* errno is set for us */
2358 if (n > 0)
2359 i += n - 1;
2360 goto nextrec;
2364 * If this is a DIF expression, and the record has a
2365 * format set, this indicates we have a CTF type name
2366 * associated with the data and we should try to print
2367 * it out by type.
2369 if (act == DTRACEACT_DIFEXPR) {
2370 const char *strdata = dt_strdata_lookup(dtp,
2371 rec->dtrd_format);
2372 if (strdata != NULL) {
2373 n = dtrace_print(dtp, fp, strdata,
2374 addr, rec->dtrd_size);
2377 * dtrace_print() will return -1 on
2378 * error, or return the number of bytes
2379 * consumed. It will return 0 if the
2380 * type couldn't be determined, and we
2381 * should fall through to the normal
2382 * trace method.
2384 if (n < 0)
2385 return (-1);
2387 if (n > 0)
2388 goto nextrec;
2392 nofmt:
2393 if (act == DTRACEACT_PRINTA) {
2394 dt_print_aggdata_t pd;
2395 dtrace_aggvarid_t *aggvars;
2396 int j, naggvars = 0;
2397 size_t size = ((epd->dtepd_nrecs - i) *
2398 sizeof (dtrace_aggvarid_t));
2400 if ((aggvars = dt_alloc(dtp, size)) == NULL)
2401 return (-1);
2404 * This might be a printa() with multiple
2405 * aggregation variables. We need to scan
2406 * forward through the records until we find
2407 * a record from a different statement.
2409 for (j = i; j < epd->dtepd_nrecs; j++) {
2410 dtrace_recdesc_t *nrec;
2411 caddr_t naddr;
2413 nrec = &epd->dtepd_rec[j];
2415 if (nrec->dtrd_uarg != rec->dtrd_uarg)
2416 break;
2418 if (nrec->dtrd_action != act) {
2419 return (dt_set_errno(dtp,
2420 EDT_BADAGG));
2423 naddr = buf->dtbd_data + offs +
2424 nrec->dtrd_offset;
2426 aggvars[naggvars++] =
2427 /* LINTED - alignment */
2428 *((dtrace_aggvarid_t *)naddr);
2431 i = j - 1;
2432 bzero(&pd, sizeof (pd));
2433 pd.dtpa_dtp = dtp;
2434 pd.dtpa_fp = fp;
2436 assert(naggvars >= 1);
2438 if (naggvars == 1) {
2439 pd.dtpa_id = aggvars[0];
2440 dt_free(dtp, aggvars);
2442 if (dt_printf(dtp, fp, "\n") < 0 ||
2443 dtrace_aggregate_walk_sorted(dtp,
2444 dt_print_agg, &pd) < 0)
2445 return (-1);
2446 goto nextrec;
2449 if (dt_printf(dtp, fp, "\n") < 0 ||
2450 dtrace_aggregate_walk_joined(dtp, aggvars,
2451 naggvars, dt_print_aggs, &pd) < 0) {
2452 dt_free(dtp, aggvars);
2453 return (-1);
2456 dt_free(dtp, aggvars);
2457 goto nextrec;
2460 if (act == DTRACEACT_TRACEMEM) {
2461 if (tracememsize == 0 ||
2462 tracememsize > rec->dtrd_size) {
2463 tracememsize = rec->dtrd_size;
2466 n = dt_print_bytes(dtp, fp, addr,
2467 tracememsize, -33, quiet, 1);
2469 tracememsize = 0;
2471 if (n < 0)
2472 return (-1);
2474 goto nextrec;
2477 switch (rec->dtrd_size) {
2478 case sizeof (uint64_t):
2479 n = dt_printf(dtp, fp,
2480 quiet ? "%lld" : " %16lld",
2481 /* LINTED - alignment */
2482 *((unsigned long long *)addr));
2483 break;
2484 case sizeof (uint32_t):
2485 n = dt_printf(dtp, fp, quiet ? "%d" : " %8d",
2486 /* LINTED - alignment */
2487 *((uint32_t *)addr));
2488 break;
2489 case sizeof (uint16_t):
2490 n = dt_printf(dtp, fp, quiet ? "%d" : " %5d",
2491 /* LINTED - alignment */
2492 *((uint16_t *)addr));
2493 break;
2494 case sizeof (uint8_t):
2495 n = dt_printf(dtp, fp, quiet ? "%d" : " %3d",
2496 *((uint8_t *)addr));
2497 break;
2498 default:
2499 n = dt_print_bytes(dtp, fp, addr,
2500 rec->dtrd_size, -33, quiet, 0);
2501 break;
2504 if (n < 0)
2505 return (-1); /* errno is set for us */
2507 nextrec:
2508 if (dt_buffered_flush(dtp, &data, rec, NULL, 0) < 0)
2509 return (-1); /* errno is set for us */
2513 * Call the record callback with a NULL record to indicate
2514 * that we're done processing this EPID.
2516 rval = (*rfunc)(&data, NULL, arg);
2517 nextepid:
2518 offs += epd->dtepd_size;
2519 dtp->dt_last_epid = id;
2520 if (just_one) {
2521 buf->dtbd_oldest = offs;
2522 break;
2526 dtp->dt_flow = data.dtpda_flow;
2527 dtp->dt_indent = data.dtpda_indent;
2528 dtp->dt_prefix = data.dtpda_prefix;
2530 if ((drops = buf->dtbd_drops) == 0)
2531 return (0);
2534 * Explicitly zero the drops to prevent us from processing them again.
2536 buf->dtbd_drops = 0;
2538 return (dt_handle_cpudrop(dtp, cpu, DTRACEDROP_PRINCIPAL, drops));
2542 * Reduce memory usage by shrinking the buffer if it's no more than half full.
2543 * Note, we need to preserve the alignment of the data at dtbd_oldest, which is
2544 * only 4-byte aligned.
2546 static void
2547 dt_realloc_buf(dtrace_hdl_t *dtp, dtrace_bufdesc_t *buf, int cursize)
2549 uint64_t used = buf->dtbd_size - buf->dtbd_oldest;
2550 if (used < cursize / 2) {
2551 int misalign = buf->dtbd_oldest & (sizeof (uint64_t) - 1);
2552 char *newdata = dt_alloc(dtp, used + misalign);
2553 if (newdata == NULL)
2554 return;
2555 bzero(newdata, misalign);
2556 bcopy(buf->dtbd_data + buf->dtbd_oldest,
2557 newdata + misalign, used);
2558 dt_free(dtp, buf->dtbd_data);
2559 buf->dtbd_oldest = misalign;
2560 buf->dtbd_size = used + misalign;
2561 buf->dtbd_data = newdata;
2566 * If the ring buffer has wrapped, the data is not in order. Rearrange it
2567 * so that it is. Note, we need to preserve the alignment of the data at
2568 * dtbd_oldest, which is only 4-byte aligned.
2570 static int
2571 dt_unring_buf(dtrace_hdl_t *dtp, dtrace_bufdesc_t *buf)
2573 int misalign;
2574 char *newdata, *ndp;
2576 if (buf->dtbd_oldest == 0)
2577 return (0);
2579 misalign = buf->dtbd_oldest & (sizeof (uint64_t) - 1);
2580 newdata = ndp = dt_alloc(dtp, buf->dtbd_size + misalign);
2582 if (newdata == NULL)
2583 return (-1);
2585 assert(0 == (buf->dtbd_size & (sizeof (uint64_t) - 1)));
2587 bzero(ndp, misalign);
2588 ndp += misalign;
2590 bcopy(buf->dtbd_data + buf->dtbd_oldest, ndp,
2591 buf->dtbd_size - buf->dtbd_oldest);
2592 ndp += buf->dtbd_size - buf->dtbd_oldest;
2594 bcopy(buf->dtbd_data, ndp, buf->dtbd_oldest);
2596 dt_free(dtp, buf->dtbd_data);
2597 buf->dtbd_oldest = 0;
2598 buf->dtbd_data = newdata;
2599 buf->dtbd_size += misalign;
2601 return (0);
2604 static void
2605 dt_put_buf(dtrace_hdl_t *dtp, dtrace_bufdesc_t *buf)
2607 dt_free(dtp, buf->dtbd_data);
2608 dt_free(dtp, buf);
2612 * Returns 0 on success, in which case *cbp will be filled in if we retrieved
2613 * data, or NULL if there is no data for this CPU.
2614 * Returns -1 on failure and sets dt_errno.
2616 static int
2617 dt_get_buf(dtrace_hdl_t *dtp, int cpu, dtrace_bufdesc_t **bufp)
2619 dtrace_optval_t size;
2620 dtrace_bufdesc_t *buf = dt_zalloc(dtp, sizeof (*buf));
2621 int error;
2623 if (buf == NULL)
2624 return (-1);
2626 (void) dtrace_getopt(dtp, "bufsize", &size);
2627 buf->dtbd_data = dt_alloc(dtp, size);
2628 if (buf->dtbd_data == NULL) {
2629 dt_free(dtp, buf);
2630 return (-1);
2632 buf->dtbd_size = size;
2633 buf->dtbd_cpu = cpu;
2635 if (dt_ioctl(dtp, DTRACEIOC_BUFSNAP, buf) == -1) {
2636 dt_put_buf(dtp, buf);
2638 * If we failed with ENOENT, it may be because the
2639 * CPU was unconfigured -- this is okay. Any other
2640 * error, however, is unexpected.
2642 if (errno == ENOENT) {
2643 *bufp = NULL;
2644 return (0);
2647 return (dt_set_errno(dtp, errno));
2650 error = dt_unring_buf(dtp, buf);
2651 if (error != 0) {
2652 dt_put_buf(dtp, buf);
2653 return (error);
2655 dt_realloc_buf(dtp, buf, size);
2657 *bufp = buf;
2658 return (0);
2661 typedef struct dt_begin {
2662 dtrace_consume_probe_f *dtbgn_probefunc;
2663 dtrace_consume_rec_f *dtbgn_recfunc;
2664 void *dtbgn_arg;
2665 dtrace_handle_err_f *dtbgn_errhdlr;
2666 void *dtbgn_errarg;
2667 int dtbgn_beginonly;
2668 } dt_begin_t;
2670 static int
2671 dt_consume_begin_probe(const dtrace_probedata_t *data, void *arg)
2673 dt_begin_t *begin = arg;
2674 dtrace_probedesc_t *pd = data->dtpda_pdesc;
2676 int r1 = (strcmp(pd->dtpd_provider, "dtrace") == 0);
2677 int r2 = (strcmp(pd->dtpd_name, "BEGIN") == 0);
2679 if (begin->dtbgn_beginonly) {
2680 if (!(r1 && r2))
2681 return (DTRACE_CONSUME_NEXT);
2682 } else {
2683 if (r1 && r2)
2684 return (DTRACE_CONSUME_NEXT);
2688 * We have a record that we're interested in. Now call the underlying
2689 * probe function...
2691 return (begin->dtbgn_probefunc(data, begin->dtbgn_arg));
2694 static int
2695 dt_consume_begin_record(const dtrace_probedata_t *data,
2696 const dtrace_recdesc_t *rec, void *arg)
2698 dt_begin_t *begin = arg;
2700 return (begin->dtbgn_recfunc(data, rec, begin->dtbgn_arg));
2703 static int
2704 dt_consume_begin_error(const dtrace_errdata_t *data, void *arg)
2706 dt_begin_t *begin = (dt_begin_t *)arg;
2707 dtrace_probedesc_t *pd = data->dteda_pdesc;
2709 int r1 = (strcmp(pd->dtpd_provider, "dtrace") == 0);
2710 int r2 = (strcmp(pd->dtpd_name, "BEGIN") == 0);
2712 if (begin->dtbgn_beginonly) {
2713 if (!(r1 && r2))
2714 return (DTRACE_HANDLE_OK);
2715 } else {
2716 if (r1 && r2)
2717 return (DTRACE_HANDLE_OK);
2720 return (begin->dtbgn_errhdlr(data, begin->dtbgn_errarg));
2723 static int
2724 dt_consume_begin(dtrace_hdl_t *dtp, FILE *fp,
2725 dtrace_consume_probe_f *pf, dtrace_consume_rec_f *rf, void *arg)
2728 * There's this idea that the BEGIN probe should be processed before
2729 * everything else, and that the END probe should be processed after
2730 * anything else. In the common case, this is pretty easy to deal
2731 * with. However, a situation may arise where the BEGIN enabling and
2732 * END enabling are on the same CPU, and some enabling in the middle
2733 * occurred on a different CPU. To deal with this (blech!) we need to
2734 * consume the BEGIN buffer up until the end of the BEGIN probe, and
2735 * then set it aside. We will then process every other CPU, and then
2736 * we'll return to the BEGIN CPU and process the rest of the data
2737 * (which will inevitably include the END probe, if any). Making this
2738 * even more complicated (!) is the library's ERROR enabling. Because
2739 * this enabling is processed before we even get into the consume call
2740 * back, any ERROR firing would result in the library's ERROR enabling
2741 * being processed twice -- once in our first pass (for BEGIN probes),
2742 * and again in our second pass (for everything but BEGIN probes). To
2743 * deal with this, we interpose on the ERROR handler to assure that we
2744 * only process ERROR enablings induced by BEGIN enablings in the
2745 * first pass, and that we only process ERROR enablings _not_ induced
2746 * by BEGIN enablings in the second pass.
2749 dt_begin_t begin;
2750 processorid_t cpu = dtp->dt_beganon;
2751 int rval, i;
2752 static int max_ncpus;
2753 dtrace_bufdesc_t *buf;
2755 dtp->dt_beganon = -1;
2757 if (dt_get_buf(dtp, cpu, &buf) != 0)
2758 return (-1);
2759 if (buf == NULL)
2760 return (0);
2762 if (!dtp->dt_stopped || buf->dtbd_cpu != dtp->dt_endedon) {
2764 * This is the simple case. We're either not stopped, or if
2765 * we are, we actually processed any END probes on another
2766 * CPU. We can simply consume this buffer and return.
2768 rval = dt_consume_cpu(dtp, fp, cpu, buf, B_FALSE,
2769 pf, rf, arg);
2770 dt_put_buf(dtp, buf);
2771 return (rval);
2774 begin.dtbgn_probefunc = pf;
2775 begin.dtbgn_recfunc = rf;
2776 begin.dtbgn_arg = arg;
2777 begin.dtbgn_beginonly = 1;
2780 * We need to interpose on the ERROR handler to be sure that we
2781 * only process ERRORs induced by BEGIN.
2783 begin.dtbgn_errhdlr = dtp->dt_errhdlr;
2784 begin.dtbgn_errarg = dtp->dt_errarg;
2785 dtp->dt_errhdlr = dt_consume_begin_error;
2786 dtp->dt_errarg = &begin;
2788 rval = dt_consume_cpu(dtp, fp, cpu, buf, B_FALSE,
2789 dt_consume_begin_probe, dt_consume_begin_record, &begin);
2791 dtp->dt_errhdlr = begin.dtbgn_errhdlr;
2792 dtp->dt_errarg = begin.dtbgn_errarg;
2794 if (rval != 0) {
2795 dt_put_buf(dtp, buf);
2796 return (rval);
2799 if (max_ncpus == 0)
2800 max_ncpus = dt_sysconf(dtp, _SC_CPUID_MAX) + 1;
2802 for (i = 0; i < max_ncpus; i++) {
2803 dtrace_bufdesc_t *nbuf;
2804 if (i == cpu)
2805 continue;
2807 if (dt_get_buf(dtp, i, &nbuf) != 0) {
2808 dt_put_buf(dtp, buf);
2809 return (-1);
2811 if (nbuf == NULL)
2812 continue;
2814 rval = dt_consume_cpu(dtp, fp, i, nbuf, B_FALSE,
2815 pf, rf, arg);
2816 dt_put_buf(dtp, nbuf);
2817 if (rval != 0) {
2818 dt_put_buf(dtp, buf);
2819 return (rval);
2824 * Okay -- we're done with the other buffers. Now we want to
2825 * reconsume the first buffer -- but this time we're looking for
2826 * everything _but_ BEGIN. And of course, in order to only consume
2827 * those ERRORs _not_ associated with BEGIN, we need to reinstall our
2828 * ERROR interposition function...
2830 begin.dtbgn_beginonly = 0;
2832 assert(begin.dtbgn_errhdlr == dtp->dt_errhdlr);
2833 assert(begin.dtbgn_errarg == dtp->dt_errarg);
2834 dtp->dt_errhdlr = dt_consume_begin_error;
2835 dtp->dt_errarg = &begin;
2837 rval = dt_consume_cpu(dtp, fp, cpu, buf, B_FALSE,
2838 dt_consume_begin_probe, dt_consume_begin_record, &begin);
2840 dtp->dt_errhdlr = begin.dtbgn_errhdlr;
2841 dtp->dt_errarg = begin.dtbgn_errarg;
2843 return (rval);
2846 /* ARGSUSED */
2847 static uint64_t
2848 dt_buf_oldest(void *elem, void *arg)
2850 dtrace_bufdesc_t *buf = elem;
2851 size_t offs = buf->dtbd_oldest;
2853 while (offs < buf->dtbd_size) {
2854 dtrace_rechdr_t *dtrh =
2855 /* LINTED - alignment */
2856 (dtrace_rechdr_t *)(buf->dtbd_data + offs);
2857 if (dtrh->dtrh_epid == DTRACE_EPIDNONE) {
2858 offs += sizeof (dtrace_epid_t);
2859 } else {
2860 return (DTRACE_RECORD_LOAD_TIMESTAMP(dtrh));
2864 /* There are no records left; use the time the buffer was retrieved. */
2865 return (buf->dtbd_timestamp);
2869 dtrace_consume(dtrace_hdl_t *dtp, FILE *fp,
2870 dtrace_consume_probe_f *pf, dtrace_consume_rec_f *rf, void *arg)
2872 dtrace_optval_t size;
2873 static int max_ncpus;
2874 int i, rval;
2875 dtrace_optval_t interval = dtp->dt_options[DTRACEOPT_SWITCHRATE];
2876 hrtime_t now = gethrtime();
2878 if (dtp->dt_lastswitch != 0) {
2879 if (now - dtp->dt_lastswitch < interval)
2880 return (0);
2882 dtp->dt_lastswitch += interval;
2883 } else {
2884 dtp->dt_lastswitch = now;
2887 if (!dtp->dt_active)
2888 return (dt_set_errno(dtp, EINVAL));
2890 if (max_ncpus == 0)
2891 max_ncpus = dt_sysconf(dtp, _SC_CPUID_MAX) + 1;
2893 if (pf == NULL)
2894 pf = (dtrace_consume_probe_f *)dt_nullprobe;
2896 if (rf == NULL)
2897 rf = (dtrace_consume_rec_f *)dt_nullrec;
2899 if (dtp->dt_options[DTRACEOPT_TEMPORAL] == DTRACEOPT_UNSET) {
2901 * The output will not be in the order it was traced. Rather,
2902 * we will consume all of the data from each CPU's buffer in
2903 * turn. We apply special handling for the records from BEGIN
2904 * and END probes so that they are consumed first and last,
2905 * respectively.
2907 * If we have just begun, we want to first process the CPU that
2908 * executed the BEGIN probe (if any).
2910 if (dtp->dt_active && dtp->dt_beganon != -1 &&
2911 (rval = dt_consume_begin(dtp, fp, pf, rf, arg)) != 0)
2912 return (rval);
2914 for (i = 0; i < max_ncpus; i++) {
2915 dtrace_bufdesc_t *buf;
2918 * If we have stopped, we want to process the CPU on
2919 * which the END probe was processed only _after_ we
2920 * have processed everything else.
2922 if (dtp->dt_stopped && (i == dtp->dt_endedon))
2923 continue;
2925 if (dt_get_buf(dtp, i, &buf) != 0)
2926 return (-1);
2927 if (buf == NULL)
2928 continue;
2930 dtp->dt_flow = 0;
2931 dtp->dt_indent = 0;
2932 dtp->dt_prefix = NULL;
2933 rval = dt_consume_cpu(dtp, fp, i,
2934 buf, B_FALSE, pf, rf, arg);
2935 dt_put_buf(dtp, buf);
2936 if (rval != 0)
2937 return (rval);
2939 if (dtp->dt_stopped) {
2940 dtrace_bufdesc_t *buf;
2942 if (dt_get_buf(dtp, dtp->dt_endedon, &buf) != 0)
2943 return (-1);
2944 if (buf == NULL)
2945 return (0);
2947 rval = dt_consume_cpu(dtp, fp, dtp->dt_endedon,
2948 buf, B_FALSE, pf, rf, arg);
2949 dt_put_buf(dtp, buf);
2950 return (rval);
2952 } else {
2954 * The output will be in the order it was traced (or for
2955 * speculations, when it was committed). We retrieve a buffer
2956 * from each CPU and put it into a priority queue, which sorts
2957 * based on the first entry in the buffer. This is sufficient
2958 * because entries within a buffer are already sorted.
2960 * We then consume records one at a time, always consuming the
2961 * oldest record, as determined by the priority queue. When
2962 * we reach the end of the time covered by these buffers,
2963 * we need to stop and retrieve more records on the next pass.
2964 * The kernel tells us the time covered by each buffer, in
2965 * dtbd_timestamp. The first buffer's timestamp tells us the
2966 * time covered by all buffers, as subsequently retrieved
2967 * buffers will cover to a more recent time.
2970 uint64_t *drops = alloca(max_ncpus * sizeof (uint64_t));
2971 uint64_t first_timestamp = 0;
2972 uint_t cookie = 0;
2973 dtrace_bufdesc_t *buf;
2975 bzero(drops, max_ncpus * sizeof (uint64_t));
2977 if (dtp->dt_bufq == NULL) {
2978 dtp->dt_bufq = dt_pq_init(dtp, max_ncpus * 2,
2979 dt_buf_oldest, NULL);
2980 if (dtp->dt_bufq == NULL) /* ENOMEM */
2981 return (-1);
2984 /* Retrieve data from each CPU. */
2985 (void) dtrace_getopt(dtp, "bufsize", &size);
2986 for (i = 0; i < max_ncpus; i++) {
2987 dtrace_bufdesc_t *buf;
2989 if (dt_get_buf(dtp, i, &buf) != 0)
2990 return (-1);
2991 if (buf != NULL) {
2992 if (first_timestamp == 0)
2993 first_timestamp = buf->dtbd_timestamp;
2994 assert(buf->dtbd_timestamp >= first_timestamp);
2996 dt_pq_insert(dtp->dt_bufq, buf);
2997 drops[i] = buf->dtbd_drops;
2998 buf->dtbd_drops = 0;
3002 /* Consume records. */
3003 for (;;) {
3004 dtrace_bufdesc_t *buf = dt_pq_pop(dtp->dt_bufq);
3005 uint64_t timestamp;
3007 if (buf == NULL)
3008 break;
3010 timestamp = dt_buf_oldest(buf, dtp);
3011 assert(timestamp >= dtp->dt_last_timestamp);
3012 dtp->dt_last_timestamp = timestamp;
3014 if (timestamp == buf->dtbd_timestamp) {
3016 * We've reached the end of the time covered
3017 * by this buffer. If this is the oldest
3018 * buffer, we must do another pass
3019 * to retrieve more data.
3021 dt_put_buf(dtp, buf);
3022 if (timestamp == first_timestamp &&
3023 !dtp->dt_stopped)
3024 break;
3025 continue;
3028 if ((rval = dt_consume_cpu(dtp, fp,
3029 buf->dtbd_cpu, buf, B_TRUE, pf, rf, arg)) != 0)
3030 return (rval);
3031 dt_pq_insert(dtp->dt_bufq, buf);
3034 /* Consume drops. */
3035 for (i = 0; i < max_ncpus; i++) {
3036 if (drops[i] != 0) {
3037 int error = dt_handle_cpudrop(dtp, i,
3038 DTRACEDROP_PRINCIPAL, drops[i]);
3039 if (error != 0)
3040 return (error);
3045 * Reduce memory usage by re-allocating smaller buffers
3046 * for the "remnants".
3048 while (buf = dt_pq_walk(dtp->dt_bufq, &cookie))
3049 dt_realloc_buf(dtp, buf, buf->dtbd_size);
3052 return (0);