1//--------------------------------------------------------------------*/
2//--- Massif: a heap profiling tool.                     ms_main.c ---*/
3//--------------------------------------------------------------------*/
4
5/*
6   This file is part of Massif, a Valgrind tool for profiling memory
7   usage of programs.
8
9   Copyright (C) 2003-2013 Nicholas Nethercote
10      njn@valgrind.org
11
12   This program is free software; you can redistribute it and/or
13   modify it under the terms of the GNU General Public License as
14   published by the Free Software Foundation; either version 2 of the
15   License, or (at your option) any later version.
16
17   This program is distributed in the hope that it will be useful, but
18   WITHOUT ANY WARRANTY; without even the implied warranty of
19   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
20   General Public License for more details.
21
22   You should have received a copy of the GNU General Public License
23   along with this program; if not, write to the Free Software
24   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
25   02111-1307, USA.
26
27   The GNU General Public License is contained in the file COPYING.
28*/
29
30//---------------------------------------------------------------------------
31// XXX:
32//---------------------------------------------------------------------------
33// Todo -- nice, but less critical:
34// - do a graph-drawing test
35// - make file format more generic.  Obstacles:
36//   - unit prefixes are not generic
37//   - preset column widths for stats are not generic
38//   - preset column headers are not generic
39//   - "Massif arguments:" line is not generic
40// - do snapshots on client requests
41//   - (Michael Meeks): have an interactive way to request a dump
42//     (callgrind_control-style)
43//     - "profile now"
44//     - "show me the extra allocations since the last snapshot"
45//     - "start/stop logging" (eg. quickly skip boring bits)
46// - Add ability to draw multiple graphs, eg. heap-only, stack-only, total.
47//   Give each graph a title.  (try to do it generically!)
48// - allow truncation of long fnnames if the exact line number is
49//   identified?  [hmm, could make getting the name of alloc-fns more
50//   difficult] [could dump full names to file, truncate in ms_print]
51// - make --show-below-main=no work
52// - Options like --alloc-fn='operator new(unsigned, std::nothrow_t const&)'
53//   don't work in a .valgrindrc file or in $VALGRIND_OPTS.
54//   m_commandline.c:add_args_from_string() needs to respect single quotes.
55// - With --stack=yes, want to add a stack trace for detailed snapshots so
56//   it's clear where/why the peak is occurring. (Mattieu Castet)  Also,
57//   possibly useful even with --stack=no? (Andi Yin)
58//
59// Performance:
60// - To run the benchmarks:
61//
62//     perl perf/vg_perf --tools=massif --reps=3 perf/{heap,tinycc} massif
63//     time valgrind --tool=massif --depth=100 konqueror
64//
65//   The other benchmarks don't do much allocation, and so give similar speeds
66//   to Nulgrind.
67//
68//   Timing results on 'nevermore' (njn's machine) as of r7013:
69//
70//     heap      0.53s  ma:12.4s (23.5x, -----)
71//     tinycc    0.46s  ma: 4.9s (10.7x, -----)
72//     many-xpts 0.08s  ma: 2.0s (25.0x, -----)
73//     konqueror 29.6s real  0:21.0s user
74//
75//   [Introduction of --time-unit=i as the default slowed things down by
76//   roughly 0--20%.]
77//
78// - get_XCon accounts for about 9% of konqueror startup time.  Try
79//   keeping XPt children sorted by 'ip' and use binary search in get_XCon.
80//   Requires factoring out binary search code from various places into a
81//   VG_(bsearch) function.
82//
83// Todo -- low priority:
84// - In each XPt, record both bytes and the number of allocations, and
85//   possibly the global number of allocations.
86// - (Andy Lin) Give a stack trace on detailed snapshots?
87// - (Artur Wisz) add a feature to Massif to ignore any heap blocks larger
88//   than a certain size!  Because: "linux's malloc allows to set a
89//   MMAP_THRESHOLD value, so we set it to 4096 - all blocks above that will
90//   be handled directly by the kernel, and are guaranteed to be returned to
91//   the system when freed. So we needed to profile only blocks below this
92//   limit."
93//
94// File format working notes:
95
96#if 0
97desc: --heap-admin=foo
98cmd: date
99time_unit: ms
100#-----------
101snapshot=0
102#-----------
103time=0
104mem_heap_B=0
105mem_heap_admin_B=0
106mem_stacks_B=0
107heap_tree=empty
108#-----------
109snapshot=1
110#-----------
111time=353
112mem_heap_B=5
113mem_heap_admin_B=0
114mem_stacks_B=0
115heap_tree=detailed
116n1: 5 (heap allocation functions) malloc/new/new[], --alloc-fns, etc.
117 n1: 5 0x27F6E0: _nl_normalize_codeset (in /lib/libc-2.3.5.so)
118  n1: 5 0x279DE6: _nl_load_locale_from_archive (in /lib/libc-2.3.5.so)
119   n1: 5 0x278E97: _nl_find_locale (in /lib/libc-2.3.5.so)
120    n1: 5 0x278871: setlocale (in /lib/libc-2.3.5.so)
121     n1: 5 0x8049821: (within /bin/date)
122      n0: 5 0x26ED5E: (below main) (in /lib/libc-2.3.5.so)
123
124
125n_events: n  time(ms)  total(B)    useful-heap(B)  admin-heap(B)  stacks(B)
126t_events: B
127n 0 0 0 0 0
128n 0 0 0 0 0
129t1: 5 <string...>
130 t1: 6 <string...>
131
132Ideas:
133- each snapshot specifies an x-axis value and one or more y-axis values.
134- can display the y-axis values separately if you like
135- can completely separate connection between snapshots and trees.
136
137Challenges:
138- how to specify and scale/abbreviate units on axes?
139- how to combine multiple values into the y-axis?
140
141--------------------------------------------------------------------------------Command:            date
142Massif arguments:   --heap-admin=foo
143ms_print arguments: massif.out
144--------------------------------------------------------------------------------
145    KB
1466.472^                                                       :#
147     |                                                       :#  ::  .    .
148     ...
149     |                                     ::@  :@    :@ :@:::#  ::  :    ::::
150   0 +-----------------------------------@---@---@-----@--@---#-------------->ms     0                                                                     713
151
152Number of snapshots: 50
153 Detailed snapshots: [2, 11, 13, 19, 25, 32 (peak)]
154--------------------------------------------------------------------------------  n       time(ms)         total(B)   useful-heap(B) admin-heap(B)    stacks(B)
155--------------------------------------------------------------------------------  0              0                0                0             0            0
156  1            345                5                5             0            0
157  2            353                5                5             0            0
158100.00% (5B) (heap allocation functions) malloc/new/new[], --alloc-fns, etc.
159->100.00% (5B) 0x27F6E0: _nl_normalize_codeset (in /lib/libc-2.3.5.so)
160#endif
161
162//---------------------------------------------------------------------------
163
164#include "pub_tool_basics.h"
165#include "pub_tool_vki.h"
166#include "pub_tool_aspacemgr.h"
167#include "pub_tool_debuginfo.h"
168#include "pub_tool_hashtable.h"
169#include "pub_tool_libcbase.h"
170#include "pub_tool_libcassert.h"
171#include "pub_tool_libcfile.h"
172#include "pub_tool_libcprint.h"
173#include "pub_tool_libcproc.h"
174#include "pub_tool_machine.h"
175#include "pub_tool_mallocfree.h"
176#include "pub_tool_options.h"
177#include "pub_tool_replacemalloc.h"
178#include "pub_tool_stacktrace.h"
179#include "pub_tool_threadstate.h"
180#include "pub_tool_tooliface.h"
181#include "pub_tool_xarray.h"
182#include "pub_tool_clientstate.h"
183#include "pub_tool_gdbserver.h"
184
185#include "pub_tool_clreq.h"           // For {MALLOC,FREE}LIKE_BLOCK
186
187//------------------------------------------------------------*/
188//--- Overview of operation                                ---*/
189//------------------------------------------------------------*/
190
191// The size of the stacks and heap is tracked.  The heap is tracked in a lot
192// of detail, enough to tell how many bytes each line of code is responsible
193// for, more or less.  The main data structure is a tree representing the
194// call tree beneath all the allocation functions like malloc().
195// (Alternatively, if --pages-as-heap=yes is specified, memory is tracked at
196// the page level, and each page is treated much like a heap block.  We use
197// "heap" throughout below to cover this case because the concepts are all the
198// same.)
199//
200// "Snapshots" are recordings of the memory usage.  There are two basic
201// kinds:
202// - Normal:  these record the current time, total memory size, total heap
203//   size, heap admin size and stack size.
204// - Detailed: these record those things in a normal snapshot, plus a very
205//   detailed XTree (see below) indicating how the heap is structured.
206//
207// Snapshots are taken every so often.  There are two storage classes of
208// snapshots:
209// - Temporary:  Massif does a temporary snapshot every so often.  The idea
210//   is to always have a certain number of temporary snapshots around.  So
211//   we take them frequently to begin with, but decreasingly often as the
212//   program continues to run.  Also, we remove some old ones after a while.
213//   Overall it's a kind of exponential decay thing.  Most of these are
214//   normal snapshots, a small fraction are detailed snapshots.
215// - Permanent:  Massif takes a permanent (detailed) snapshot in some
216//   circumstances.  They are:
217//   - Peak snapshot:  When the memory usage peak is reached, it takes a
218//     snapshot.  It keeps this, unless the peak is subsequently exceeded,
219//     in which case it will overwrite the peak snapshot.
220//   - User-requested snapshots:  These are done in response to client
221//     requests.  They are always kept.
222
223// Used for printing things when clo_verbosity > 1.
224#define VERB(verb, format, args...) \
225   if (VG_(clo_verbosity) > verb) { \
226      VG_(dmsg)("Massif: " format, ##args); \
227   }
228
229//------------------------------------------------------------//
230//--- Statistics                                           ---//
231//------------------------------------------------------------//
232
233// Konqueror startup, to give an idea of the numbers involved with a biggish
234// program, with default depth:
235//
236//  depth=3                   depth=40
237//  - 310,000 allocations
238//  - 300,000 frees
239//  -  15,000 XPts            800,000 XPts
240//  -   1,800 top-XPts
241
242static UInt n_heap_allocs           = 0;
243static UInt n_heap_reallocs         = 0;
244static UInt n_heap_frees            = 0;
245static UInt n_ignored_heap_allocs   = 0;
246static UInt n_ignored_heap_frees    = 0;
247static UInt n_ignored_heap_reallocs = 0;
248static UInt n_stack_allocs          = 0;
249static UInt n_stack_frees           = 0;
250static UInt n_xpts                  = 0;
251static UInt n_xpt_init_expansions   = 0;
252static UInt n_xpt_later_expansions  = 0;
253static UInt n_sxpt_allocs           = 0;
254static UInt n_sxpt_frees            = 0;
255static UInt n_skipped_snapshots     = 0;
256static UInt n_real_snapshots        = 0;
257static UInt n_detailed_snapshots    = 0;
258static UInt n_peak_snapshots        = 0;
259static UInt n_cullings              = 0;
260static UInt n_XCon_redos            = 0;
261
262//------------------------------------------------------------//
263//--- Globals                                              ---//
264//------------------------------------------------------------//
265
266// Number of guest instructions executed so far.  Only used with
267// --time-unit=i.
268static Long guest_instrs_executed = 0;
269
270static SizeT heap_szB       = 0; // Live heap size
271static SizeT heap_extra_szB = 0; // Live heap extra size -- slop + admin bytes
272static SizeT stacks_szB     = 0; // Live stacks size
273
274// This is the total size from the current peak snapshot, or 0 if no peak
275// snapshot has been taken yet.
276static SizeT peak_snapshot_total_szB = 0;
277
278// Incremented every time memory is allocated/deallocated, by the
279// allocated/deallocated amount;  includes heap, heap-admin and stack
280// memory.  An alternative to milliseconds as a unit of program "time".
281static ULong total_allocs_deallocs_szB = 0;
282
283// When running with --heap=yes --pages-as-heap=no, we don't start taking
284// snapshots until the first basic block is executed, rather than doing it in
285// ms_post_clo_init (which is the obvious spot), for two reasons.
286// - It lets us ignore stack events prior to that, because they're not
287//   really proper ones and just would screw things up.
288// - Because there's still some core initialisation to do, and so there
289//   would be an artificial time gap between the first and second snapshots.
290//
291// When running with --heap=yes --pages-as-heap=yes, snapshots start much
292// earlier due to new_mem_startup so this isn't relevant.
293//
294static Bool have_started_executing_code = False;
295
296//------------------------------------------------------------//
297//--- Alloc fns                                            ---//
298//------------------------------------------------------------//
299
300static XArray* alloc_fns;
301static XArray* ignore_fns;
302
303static void init_alloc_fns(void)
304{
305   // Create the list, and add the default elements.
306   alloc_fns = VG_(newXA)(VG_(malloc), "ms.main.iaf.1",
307                                       VG_(free), sizeof(HChar*));
308   #define DO(x)  { const HChar* s = x; VG_(addToXA)(alloc_fns, &s); }
309
310   // Ordered roughly according to (presumed) frequency.
311   // Nb: The C++ "operator new*" ones are overloadable.  We include them
312   // always anyway, because even if they're overloaded, it would be a
313   // prodigiously stupid overloading that caused them to not allocate
314   // memory.
315   //
316   // XXX: because we don't look at the first stack entry (unless it's a
317   // custom allocation) there's not much point to having all these alloc
318   // functions here -- they should never appear anywhere (I think?) other
319   // than the top stack entry.  The only exceptions are those that in
320   // vg_replace_malloc.c are partly or fully implemented in terms of another
321   // alloc function: realloc (which uses malloc);  valloc,
322   // malloc_zone_valloc, posix_memalign and memalign_common (which use
323   // memalign).
324   //
325   DO("malloc"                                              );
326   DO("__builtin_new"                                       );
327   DO("operator new(unsigned)"                              );
328   DO("operator new(unsigned long)"                         );
329   DO("__builtin_vec_new"                                   );
330   DO("operator new[](unsigned)"                            );
331   DO("operator new[](unsigned long)"                       );
332   DO("calloc"                                              );
333   DO("realloc"                                             );
334   DO("memalign"                                            );
335   DO("posix_memalign"                                      );
336   DO("valloc"                                              );
337   DO("operator new(unsigned, std::nothrow_t const&)"       );
338   DO("operator new[](unsigned, std::nothrow_t const&)"     );
339   DO("operator new(unsigned long, std::nothrow_t const&)"  );
340   DO("operator new[](unsigned long, std::nothrow_t const&)");
341#if defined(VGO_darwin)
342   DO("malloc_zone_malloc"                                  );
343   DO("malloc_zone_calloc"                                  );
344   DO("malloc_zone_realloc"                                 );
345   DO("malloc_zone_memalign"                                );
346   DO("malloc_zone_valloc"                                  );
347#endif
348}
349
350static void init_ignore_fns(void)
351{
352   // Create the (empty) list.
353   ignore_fns = VG_(newXA)(VG_(malloc), "ms.main.iif.1",
354                                        VG_(free), sizeof(HChar*));
355}
356
357// Determines if the named function is a member of the XArray.
358static Bool is_member_fn(XArray* fns, const HChar* fnname)
359{
360   HChar** fn_ptr;
361   Int i;
362
363   // Nb: It's a linear search through the list, because we're comparing
364   // strings rather than pointers to strings.
365   // Nb: This gets called a lot.  It was an OSet, but they're quite slow to
366   // iterate through so it wasn't a good choice.
367   for (i = 0; i < VG_(sizeXA)(fns); i++) {
368      fn_ptr = VG_(indexXA)(fns, i);
369      if (VG_STREQ(fnname, *fn_ptr))
370         return True;
371   }
372   return False;
373}
374
375
376//------------------------------------------------------------//
377//--- Command line args                                    ---//
378//------------------------------------------------------------//
379
380#define MAX_DEPTH       200
381
382typedef enum { TimeI, TimeMS, TimeB } TimeUnit;
383
384static const HChar* TimeUnit_to_string(TimeUnit time_unit)
385{
386   switch (time_unit) {
387   case TimeI:  return "i";
388   case TimeMS: return "ms";
389   case TimeB:  return "B";
390   default:     tl_assert2(0, "TimeUnit_to_string: unrecognised TimeUnit");
391   }
392}
393
394static Bool   clo_heap            = True;
395   // clo_heap_admin is deliberately a word-sized type.  At one point it was
396   // a UInt, but this caused problems on 64-bit machines when it was
397   // multiplied by a small negative number and then promoted to a
398   // word-sized type -- it ended up with a value of 4.2 billion.  Sigh.
399static SSizeT clo_heap_admin      = 8;
400static Bool   clo_pages_as_heap   = False;
401static Bool   clo_stacks          = False;
402static Int    clo_depth           = 30;
403static double clo_threshold       = 1.0;  // percentage
404static double clo_peak_inaccuracy = 1.0;  // percentage
405static Int    clo_time_unit       = TimeI;
406static Int    clo_detailed_freq   = 10;
407static Int    clo_max_snapshots   = 100;
408static const HChar* clo_massif_out_file = "massif.out.%p";
409
410static XArray* args_for_massif;
411
412static Bool ms_process_cmd_line_option(const HChar* arg)
413{
414   const HChar* tmp_str;
415
416   // Remember the arg for later use.
417   VG_(addToXA)(args_for_massif, &arg);
418
419        if VG_BOOL_CLO(arg, "--heap",           clo_heap)   {}
420   else if VG_BINT_CLO(arg, "--heap-admin",     clo_heap_admin, 0, 1024) {}
421
422   else if VG_BOOL_CLO(arg, "--stacks",         clo_stacks) {}
423
424   else if VG_BOOL_CLO(arg, "--pages-as-heap",  clo_pages_as_heap) {}
425
426   else if VG_BINT_CLO(arg, "--depth",          clo_depth, 1, MAX_DEPTH) {}
427
428   else if VG_STR_CLO(arg, "--alloc-fn",        tmp_str) {
429      VG_(addToXA)(alloc_fns, &tmp_str);
430   }
431   else if VG_STR_CLO(arg, "--ignore-fn",       tmp_str) {
432      VG_(addToXA)(ignore_fns, &tmp_str);
433   }
434
435   else if VG_DBL_CLO(arg, "--threshold",  clo_threshold) {
436      if (clo_threshold < 0 || clo_threshold > 100) {
437         VG_(fmsg_bad_option)(arg,
438            "--threshold must be between 0.0 and 100.0\n");
439      }
440   }
441
442   else if VG_DBL_CLO(arg, "--peak-inaccuracy", clo_peak_inaccuracy) {}
443
444   else if VG_XACT_CLO(arg, "--time-unit=i",    clo_time_unit, TimeI)  {}
445   else if VG_XACT_CLO(arg, "--time-unit=ms",   clo_time_unit, TimeMS) {}
446   else if VG_XACT_CLO(arg, "--time-unit=B",    clo_time_unit, TimeB)  {}
447
448   else if VG_BINT_CLO(arg, "--detailed-freq",  clo_detailed_freq, 1, 1000000) {}
449
450   else if VG_BINT_CLO(arg, "--max-snapshots",  clo_max_snapshots, 10, 1000) {}
451
452   else if VG_STR_CLO(arg, "--massif-out-file", clo_massif_out_file) {}
453
454   else
455      return VG_(replacement_malloc_process_cmd_line_option)(arg);
456
457   return True;
458}
459
460static void ms_print_usage(void)
461{
462   VG_(printf)(
463"    --heap=no|yes             profile heap blocks [yes]\n"
464"    --heap-admin=<size>       average admin bytes per heap block;\n"
465"                               ignored if --heap=no [8]\n"
466"    --stacks=no|yes           profile stack(s) [no]\n"
467"    --pages-as-heap=no|yes    profile memory at the page level [no]\n"
468"    --depth=<number>          depth of contexts [30]\n"
469"    --alloc-fn=<name>         specify <name> as an alloc function [empty]\n"
470"    --ignore-fn=<name>        ignore heap allocations within <name> [empty]\n"
471"    --threshold=<m.n>         significance threshold, as a percentage [1.0]\n"
472"    --peak-inaccuracy=<m.n>   maximum peak inaccuracy, as a percentage [1.0]\n"
473"    --time-unit=i|ms|B        time unit: instructions executed, milliseconds\n"
474"                              or heap bytes alloc'd/dealloc'd [i]\n"
475"    --detailed-freq=<N>       every Nth snapshot should be detailed [10]\n"
476"    --max-snapshots=<N>       maximum number of snapshots recorded [100]\n"
477"    --massif-out-file=<file>  output file name [massif.out.%%p]\n"
478   );
479}
480
481static void ms_print_debug_usage(void)
482{
483   VG_(printf)(
484"    (none)\n"
485   );
486}
487
488
489//------------------------------------------------------------//
490//--- XPts, XTrees and XCons                               ---//
491//------------------------------------------------------------//
492
493// An XPt represents an "execution point", ie. a code address.  Each XPt is
494// part of a tree of XPts (an "execution tree", or "XTree").  The details of
495// the heap are represented by a single XTree.
496//
497// The root of the tree is 'alloc_xpt', which represents all allocation
498// functions, eg:
499// - malloc/calloc/realloc/memalign/new/new[];
500// - user-specified allocation functions (using --alloc-fn);
501// - custom allocation (MALLOCLIKE) points
502// It's a bit of a fake XPt (ie. its 'ip' is zero), and is only used because
503// it makes the code simpler.
504//
505// Any child of 'alloc_xpt' is called a "top-XPt".  The XPts at the bottom
506// of an XTree (leaf nodes) are "bottom-XPTs".
507//
508// Each path from a top-XPt to a bottom-XPt through an XTree gives an
509// execution context ("XCon"), ie. a stack trace.  (And sub-paths represent
510// stack sub-traces.)  The number of XCons in an XTree is equal to the
511// number of bottom-XPTs in that XTree.
512//
513//      alloc_xpt       XTrees are bi-directional.
514//        | ^
515//        v |
516//     > parent <       Example: if child1() calls parent() and child2()
517//    /    |     \      also calls parent(), and parent() calls malloc(),
518//   |    / \     |     the XTree will look like this.
519//   |   v   v    |
520//  child1   child2
521//
522// (Note that malformed stack traces can lead to difficulties.  See the
523// comment at the bottom of get_XCon.)
524//
525// XTrees and XPts are mirrored by SXTrees and SXPts, where the 'S' is short
526// for "saved".  When the XTree is duplicated for a snapshot, we duplicate
527// it as an SXTree, which is similar but omits some things it does not need,
528// and aggregates up insignificant nodes.  This is important as an SXTree is
529// typically much smaller than an XTree.
530
531// XXX: make XPt and SXPt extensible arrays, to avoid having to do two
532// allocations per Pt.
533
534typedef struct _XPt XPt;
535struct _XPt {
536   Addr  ip;              // code address
537
538   // Bottom-XPts: space for the precise context.
539   // Other XPts:  space of all the descendent bottom-XPts.
540   // Nb: this value goes up and down as the program executes.
541   SizeT szB;
542
543   XPt*  parent;           // pointer to parent XPt
544
545   // Children.
546   // n_children and max_children are 32-bit integers.  16-bit integers
547   // are too small -- a very big program might have more than 65536
548   // allocation points (ie. top-XPts) -- Konqueror starting up has 1800.
549   UInt  n_children;       // number of children
550   UInt  max_children;     // capacity of children array
551   XPt** children;         // pointers to children XPts
552};
553
554typedef
555   enum {
556      SigSXPt,
557      InsigSXPt
558   }
559   SXPtTag;
560
561typedef struct _SXPt SXPt;
562struct _SXPt {
563   SXPtTag tag;
564   SizeT szB;              // memory size for the node, be it Sig or Insig
565   union {
566      // An SXPt representing a single significant code location.  Much like
567      // an XPt, minus the fields that aren't necessary.
568      struct {
569         Addr   ip;
570         UInt   n_children;
571         SXPt** children;
572      }
573      Sig;
574
575      // An SXPt representing one or more code locations, all below the
576      // significance threshold.
577      struct {
578         Int   n_xpts;     // number of aggregated XPts
579      }
580      Insig;
581   };
582};
583
584// Fake XPt representing all allocation functions like malloc().  Acts as
585// parent node to all top-XPts.
586static XPt* alloc_xpt;
587
588static XPt* new_XPt(Addr ip, XPt* parent)
589{
590   // XPts are never freed, so we can use VG_(perm_malloc) to allocate them.
591   // Note that we cannot use VG_(perm_malloc) for the 'children' array, because
592   // that needs to be resizable.
593   XPt* xpt    = VG_(perm_malloc)(sizeof(XPt), vg_alignof(XPt));
594   xpt->ip     = ip;
595   xpt->szB    = 0;
596   xpt->parent = parent;
597
598   // We don't initially allocate any space for children.  We let that
599   // happen on demand.  Many XPts (ie. all the bottom-XPts) don't have any
600   // children anyway.
601   xpt->n_children   = 0;
602   xpt->max_children = 0;
603   xpt->children     = NULL;
604
605   // Update statistics
606   n_xpts++;
607
608   return xpt;
609}
610
611static void add_child_xpt(XPt* parent, XPt* child)
612{
613   // Expand 'children' if necessary.
614   tl_assert(parent->n_children <= parent->max_children);
615   if (parent->n_children == parent->max_children) {
616      if (0 == parent->max_children) {
617         parent->max_children = 4;
618         parent->children = VG_(malloc)( "ms.main.acx.1",
619                                         parent->max_children * sizeof(XPt*) );
620         n_xpt_init_expansions++;
621      } else {
622         parent->max_children *= 2;    // Double size
623         parent->children = VG_(realloc)( "ms.main.acx.2",
624                                          parent->children,
625                                          parent->max_children * sizeof(XPt*) );
626         n_xpt_later_expansions++;
627      }
628   }
629
630   // Insert new child XPt in parent's children list.
631   parent->children[ parent->n_children++ ] = child;
632}
633
634// Reverse comparison for a reverse sort -- biggest to smallest.
635static Int SXPt_revcmp_szB(const void* n1, const void* n2)
636{
637   const SXPt* sxpt1 = *(const SXPt *const *)n1;
638   const SXPt* sxpt2 = *(const SXPt *const *)n2;
639   return ( sxpt1->szB < sxpt2->szB ?  1
640          : sxpt1->szB > sxpt2->szB ? -1
641          :                            0);
642}
643
644//------------------------------------------------------------//
645//--- XTree Operations                                     ---//
646//------------------------------------------------------------//
647
648// Duplicates an XTree as an SXTree.
649static SXPt* dup_XTree(XPt* xpt, SizeT total_szB)
650{
651   Int  i, n_sig_children, n_insig_children, n_child_sxpts;
652   SizeT sig_child_threshold_szB;
653   SXPt* sxpt;
654
655   // Number of XPt children  Action for SXPT
656   // ------------------      ---------------
657   // 0 sig, 0 insig          alloc 0 children
658   // N sig, 0 insig          alloc N children, dup all
659   // N sig, M insig          alloc N+1, dup first N, aggregate remaining M
660   // 0 sig, M insig          alloc 1, aggregate M
661
662   // Work out how big a child must be to be significant.  If the current
663   // total_szB is zero, then we set it to 1, which means everything will be
664   // judged insignificant -- this is sensible, as there's no point showing
665   // any detail for this case.  Unless they used --threshold=0, in which
666   // case we show them everything because that's what they asked for.
667   //
668   // Nb: We do this once now, rather than once per child, because if we do
669   // that the cost of all the divisions adds up to something significant.
670   if (0 == total_szB && 0 != clo_threshold) {
671      sig_child_threshold_szB = 1;
672   } else {
673      sig_child_threshold_szB = (SizeT)((total_szB * clo_threshold) / 100);
674   }
675
676   // How many children are significant?  And do we need an aggregate SXPt?
677   n_sig_children = 0;
678   for (i = 0; i < xpt->n_children; i++) {
679      if (xpt->children[i]->szB >= sig_child_threshold_szB) {
680         n_sig_children++;
681      }
682   }
683   n_insig_children = xpt->n_children - n_sig_children;
684   n_child_sxpts = n_sig_children + ( n_insig_children > 0 ? 1 : 0 );
685
686   // Duplicate the XPt.
687   sxpt                 = VG_(malloc)("ms.main.dX.1", sizeof(SXPt));
688   n_sxpt_allocs++;
689   sxpt->tag            = SigSXPt;
690   sxpt->szB            = xpt->szB;
691   sxpt->Sig.ip         = xpt->ip;
692   sxpt->Sig.n_children = n_child_sxpts;
693
694   // Create the SXPt's children.
695   if (n_child_sxpts > 0) {
696      Int j;
697      SizeT sig_children_szB = 0, insig_children_szB = 0;
698      sxpt->Sig.children = VG_(malloc)("ms.main.dX.2",
699                                       n_child_sxpts * sizeof(SXPt*));
700
701      // Duplicate the significant children.  (Nb: sig_children_szB +
702      // insig_children_szB doesn't necessarily equal xpt->szB.)
703      j = 0;
704      for (i = 0; i < xpt->n_children; i++) {
705         if (xpt->children[i]->szB >= sig_child_threshold_szB) {
706            sxpt->Sig.children[j++] = dup_XTree(xpt->children[i], total_szB);
707            sig_children_szB   += xpt->children[i]->szB;
708         } else {
709            insig_children_szB += xpt->children[i]->szB;
710         }
711      }
712
713      // Create the SXPt for the insignificant children, if any, and put it
714      // in the last child entry.
715      if (n_insig_children > 0) {
716         // Nb: We 'n_sxpt_allocs' here because creating an Insig SXPt
717         // doesn't involve a call to dup_XTree().
718         SXPt* insig_sxpt = VG_(malloc)("ms.main.dX.3", sizeof(SXPt));
719         n_sxpt_allocs++;
720         insig_sxpt->tag = InsigSXPt;
721         insig_sxpt->szB = insig_children_szB;
722         insig_sxpt->Insig.n_xpts = n_insig_children;
723         sxpt->Sig.children[n_sig_children] = insig_sxpt;
724      }
725   } else {
726      sxpt->Sig.children = NULL;
727   }
728
729   return sxpt;
730}
731
732static void free_SXTree(SXPt* sxpt)
733{
734   Int  i;
735   tl_assert(sxpt != NULL);
736
737   switch (sxpt->tag) {
738    case SigSXPt:
739      // Free all children SXPts, then the children array.
740      for (i = 0; i < sxpt->Sig.n_children; i++) {
741         free_SXTree(sxpt->Sig.children[i]);
742         sxpt->Sig.children[i] = NULL;
743      }
744      VG_(free)(sxpt->Sig.children);  sxpt->Sig.children = NULL;
745      break;
746
747    case InsigSXPt:
748      break;
749
750    default: tl_assert2(0, "free_SXTree: unknown SXPt tag");
751   }
752
753   // Free the SXPt itself.
754   VG_(free)(sxpt);     sxpt = NULL;
755   n_sxpt_frees++;
756}
757
758// Sanity checking:  we periodically check the heap XTree with
759// ms_expensive_sanity_check.
760static void sanity_check_XTree(XPt* xpt, XPt* parent)
761{
762   tl_assert(xpt != NULL);
763
764   // Check back-pointer.
765   tl_assert2(xpt->parent == parent,
766      "xpt->parent = %p, parent = %p\n", xpt->parent, parent);
767
768   // Check children counts look sane.
769   tl_assert(xpt->n_children <= xpt->max_children);
770
771   // Unfortunately, xpt's size is not necessarily equal to the sum of xpt's
772   // children's sizes.  See comment at the bottom of get_XCon.
773}
774
775// Sanity checking:  we check SXTrees (which are in snapshots) after
776// snapshots are created, before they are deleted, and before they are
777// printed.
778static void sanity_check_SXTree(SXPt* sxpt)
779{
780   Int i;
781
782   tl_assert(sxpt != NULL);
783
784   // Check the sum of any children szBs equals the SXPt's szB.  Check the
785   // children at the same time.
786   switch (sxpt->tag) {
787    case SigSXPt: {
788      if (sxpt->Sig.n_children > 0) {
789         for (i = 0; i < sxpt->Sig.n_children; i++) {
790            sanity_check_SXTree(sxpt->Sig.children[i]);
791         }
792      }
793      break;
794    }
795    case InsigSXPt:
796      break;         // do nothing
797
798    default: tl_assert2(0, "sanity_check_SXTree: unknown SXPt tag");
799   }
800}
801
802
803//------------------------------------------------------------//
804//--- XCon Operations                                      ---//
805//------------------------------------------------------------//
806
807// This is the limit on the number of removed alloc-fns that can be in a
808// single XCon.
809#define MAX_OVERESTIMATE   50
810#define MAX_IPS            (MAX_DEPTH + MAX_OVERESTIMATE)
811
812// This is used for various buffers which can hold function names/IP
813// description.  Some C++ names can get really long so 1024 isn't big
814// enough.
815#define BUF_LEN   2048
816
817// Determine if the given IP belongs to a function that should be ignored.
818static Bool fn_should_be_ignored(Addr ip)
819{
820   static HChar buf[BUF_LEN];
821   return
822      ( VG_(get_fnname)(ip, buf, BUF_LEN) && is_member_fn(ignore_fns, buf)
823      ? True : False );
824}
825
826// Get the stack trace for an XCon, filtering out uninteresting entries:
827// alloc-fns and entries above alloc-fns, and entries below main-or-below-main.
828//   Eg:       alloc-fn1 / alloc-fn2 / a / b / main / (below main) / c
829//   becomes:  a / b / main
830// Nb: it's possible to end up with an empty trace, eg. if 'main' is marked
831// as an alloc-fn.  This is ok.
832static
833Int get_IPs( ThreadId tid, Bool exclude_first_entry, Addr ips[])
834{
835   static HChar buf[BUF_LEN];
836   Int n_ips, i, n_alloc_fns_removed;
837   Int overestimate;
838   Bool redo;
839
840   // We ask for a few more IPs than clo_depth suggests we need.  Then we
841   // remove every entry that is an alloc-fn.  Depending on the
842   // circumstances, we may need to redo it all, asking for more IPs.
843   // Details:
844   // - If the original stack trace is smaller than asked-for, redo=False
845   // - Else if after filtering we have >= clo_depth IPs,      redo=False
846   // - Else redo=True
847   // In other words, to redo, we'd have to get a stack trace as big as we
848   // asked for and remove more than 'overestimate' alloc-fns.
849
850   // Main loop.
851   redo = True;      // Assume this to begin with.
852   for (overestimate = 3; redo; overestimate += 6) {
853      // This should never happen -- would require MAX_OVERESTIMATE
854      // alloc-fns to be removed from the stack trace.
855      if (overestimate > MAX_OVERESTIMATE)
856         VG_(tool_panic)("get_IPs: ips[] too small, inc. MAX_OVERESTIMATE?");
857
858      // Ask for more IPs than clo_depth suggests we need.
859      n_ips = VG_(get_StackTrace)( tid, ips, clo_depth + overestimate,
860                                   NULL/*array to dump SP values in*/,
861                                   NULL/*array to dump FP values in*/,
862                                   0/*first_ip_delta*/ );
863      tl_assert(n_ips > 0);
864
865      // If the original stack trace is smaller than asked-for, redo=False.
866      if (n_ips < clo_depth + overestimate) { redo = False; }
867
868      // Filter out alloc fns.  If requested, we automatically remove the
869      // first entry (which presumably will be something like malloc or
870      // __builtin_new that we're sure to filter out) without looking at it,
871      // because VG_(get_fnname) is expensive.
872      n_alloc_fns_removed = ( exclude_first_entry ? 1 : 0 );
873      for (i = n_alloc_fns_removed; i < n_ips; i++) {
874         if (VG_(get_fnname)(ips[i], buf, BUF_LEN)) {
875            if (is_member_fn(alloc_fns, buf)) {
876               n_alloc_fns_removed++;
877            } else {
878               break;
879            }
880         }
881      }
882      // Remove the alloc fns by shuffling the rest down over them.
883      n_ips -= n_alloc_fns_removed;
884      for (i = 0; i < n_ips; i++) {
885         ips[i] = ips[i + n_alloc_fns_removed];
886      }
887
888      // If after filtering we have >= clo_depth IPs, redo=False
889      if (n_ips >= clo_depth) {
890         redo = False;
891         n_ips = clo_depth;      // Ignore any IPs below --depth.
892      }
893
894      if (redo) {
895         n_XCon_redos++;
896      }
897   }
898   return n_ips;
899}
900
901// Gets an XCon and puts it in the tree.  Returns the XCon's bottom-XPt.
902// Unless the allocation should be ignored, in which case we return NULL.
903static XPt* get_XCon( ThreadId tid, Bool exclude_first_entry )
904{
905   static Addr ips[MAX_IPS];
906   Int i;
907   XPt* xpt = alloc_xpt;
908
909   // After this call, the IPs we want are in ips[0]..ips[n_ips-1].
910   Int n_ips = get_IPs(tid, exclude_first_entry, ips);
911
912   // Should we ignore this allocation?  (Nb: n_ips can be zero, eg. if
913   // 'main' is marked as an alloc-fn.)
914   if (n_ips > 0 && fn_should_be_ignored(ips[0])) {
915      return NULL;
916   }
917
918   // Now do the search/insertion of the XCon.
919   for (i = 0; i < n_ips; i++) {
920      Addr ip = ips[i];
921      Int ch;
922      // Look for IP in xpt's children.
923      // Linear search, ugh -- about 10% of time for konqueror startup tried
924      // caching last result, only hit about 4% for konqueror.
925      // Nb:  this search hits about 98% of the time for konqueror
926      for (ch = 0; True; ch++) {
927         if (ch == xpt->n_children) {
928            // IP not found in the children.
929            // Create and add new child XPt, then stop.
930            XPt* new_child_xpt = new_XPt(ip, xpt);
931            add_child_xpt(xpt, new_child_xpt);
932            xpt = new_child_xpt;
933            break;
934
935         } else if (ip == xpt->children[ch]->ip) {
936            // Found the IP in the children, stop.
937            xpt = xpt->children[ch];
938            break;
939         }
940      }
941   }
942
943   // [Note: several comments refer to this comment.  Do not delete it
944   // without updating them.]
945   //
946   // A complication... If all stack traces were well-formed, then the
947   // returned xpt would always be a bottom-XPt.  As a consequence, an XPt's
948   // size would always be equal to the sum of its children's sizes, which
949   // is an excellent sanity check.
950   //
951   // Unfortunately, stack traces occasionally are malformed, ie. truncated.
952   // This allows a stack trace to be a sub-trace of another, eg. a/b/c is a
953   // sub-trace of a/b/c/d.  So we can't assume this xpt is a bottom-XPt;
954   // nor can we do sanity check an XPt's size against its children's sizes.
955   // This is annoying, but must be dealt with.  (Older versions of Massif
956   // had this assertion in, and it was reported to fail by real users a
957   // couple of times.)  Even more annoyingly, I can't come up with a simple
958   // test case that exhibit such a malformed stack trace, so I can't
959   // regression test it.  Sigh.
960   //
961   // However, we can print a warning, so that if it happens (unexpectedly)
962   // in existing regression tests we'll know.  Also, it warns users that
963   // the output snapshots may not add up the way they might expect.
964   //
965   //tl_assert(0 == xpt->n_children); // Must be bottom-XPt
966   if (0 != xpt->n_children) {
967      static Int n_moans = 0;
968      if (n_moans < 3) {
969         VG_(umsg)(
970            "Warning: Malformed stack trace detected.  In Massif's output,\n");
971         VG_(umsg)(
972            "         the size of an entry's child entries may not sum up\n");
973         VG_(umsg)(
974            "         to the entry's size as they normally do.\n");
975         n_moans++;
976         if (3 == n_moans)
977            VG_(umsg)(
978            "         (And Massif now won't warn about this again.)\n");
979      }
980   }
981   return xpt;
982}
983
984// Update 'szB' of every XPt in the XCon, by percolating upwards.
985static void update_XCon(XPt* xpt, SSizeT space_delta)
986{
987   tl_assert(clo_heap);
988   tl_assert(NULL != xpt);
989
990   if (0 == space_delta)
991      return;
992
993   while (xpt != alloc_xpt) {
994      if (space_delta < 0) tl_assert(xpt->szB >= -space_delta);
995      xpt->szB += space_delta;
996      xpt = xpt->parent;
997   }
998   if (space_delta < 0) tl_assert(alloc_xpt->szB >= -space_delta);
999   alloc_xpt->szB += space_delta;
1000}
1001
1002
1003//------------------------------------------------------------//
1004//--- Snapshots                                            ---//
1005//------------------------------------------------------------//
1006
1007// Snapshots are done in a way so that we always have a reasonable number of
1008// them.  We start by taking them quickly.  Once we hit our limit, we cull
1009// some (eg. half), and start taking them more slowly.  Once we hit the
1010// limit again, we again cull and then take them even more slowly, and so
1011// on.
1012
1013// Time is measured either in i or ms or bytes, depending on the --time-unit
1014// option.  It's a Long because it can exceed 32-bits reasonably easily, and
1015// because we need to allow negative values to represent unset times.
1016typedef Long Time;
1017
1018#define UNUSED_SNAPSHOT_TIME  -333  // A conspicuous negative number.
1019
1020typedef
1021   enum {
1022      Normal = 77,
1023      Peak,
1024      Unused
1025   }
1026   SnapshotKind;
1027
1028typedef
1029   struct {
1030      SnapshotKind kind;
1031      Time  time;
1032      SizeT heap_szB;
1033      SizeT heap_extra_szB;// Heap slop + admin bytes.
1034      SizeT stacks_szB;
1035      SXPt* alloc_sxpt;    // Heap XTree root, if a detailed snapshot,
1036   }                       // otherwise NULL.
1037   Snapshot;
1038
1039static UInt      next_snapshot_i = 0;  // Index of where next snapshot will go.
1040static Snapshot* snapshots;            // Array of snapshots.
1041
1042static Bool is_snapshot_in_use(Snapshot* snapshot)
1043{
1044   if (Unused == snapshot->kind) {
1045      // If snapshot is unused, check all the fields are unset.
1046      tl_assert(snapshot->time           == UNUSED_SNAPSHOT_TIME);
1047      tl_assert(snapshot->heap_extra_szB == 0);
1048      tl_assert(snapshot->heap_szB       == 0);
1049      tl_assert(snapshot->stacks_szB     == 0);
1050      tl_assert(snapshot->alloc_sxpt     == NULL);
1051      return False;
1052   } else {
1053      tl_assert(snapshot->time           != UNUSED_SNAPSHOT_TIME);
1054      return True;
1055   }
1056}
1057
1058static Bool is_detailed_snapshot(Snapshot* snapshot)
1059{
1060   return (snapshot->alloc_sxpt ? True : False);
1061}
1062
1063static Bool is_uncullable_snapshot(Snapshot* snapshot)
1064{
1065   return &snapshots[0] == snapshot                   // First snapshot
1066       || &snapshots[next_snapshot_i-1] == snapshot   // Last snapshot
1067       || snapshot->kind == Peak;                     // Peak snapshot
1068}
1069
1070static void sanity_check_snapshot(Snapshot* snapshot)
1071{
1072   if (snapshot->alloc_sxpt) {
1073      sanity_check_SXTree(snapshot->alloc_sxpt);
1074   }
1075}
1076
1077// All the used entries should look used, all the unused ones should be clear.
1078static void sanity_check_snapshots_array(void)
1079{
1080   Int i;
1081   for (i = 0; i < next_snapshot_i; i++) {
1082      tl_assert( is_snapshot_in_use( & snapshots[i] ));
1083   }
1084   for (    ; i < clo_max_snapshots; i++) {
1085      tl_assert(!is_snapshot_in_use( & snapshots[i] ));
1086   }
1087}
1088
1089// This zeroes all the fields in the snapshot, but does not free the heap
1090// XTree if present.  It also does a sanity check unless asked not to;  we
1091// can't sanity check at startup when clearing the initial snapshots because
1092// they're full of junk.
1093static void clear_snapshot(Snapshot* snapshot, Bool do_sanity_check)
1094{
1095   if (do_sanity_check) sanity_check_snapshot(snapshot);
1096   snapshot->kind           = Unused;
1097   snapshot->time           = UNUSED_SNAPSHOT_TIME;
1098   snapshot->heap_extra_szB = 0;
1099   snapshot->heap_szB       = 0;
1100   snapshot->stacks_szB     = 0;
1101   snapshot->alloc_sxpt     = NULL;
1102}
1103
1104// This zeroes all the fields in the snapshot, and frees the heap XTree if
1105// present.
1106static void delete_snapshot(Snapshot* snapshot)
1107{
1108   // Nb: if there's an XTree, we free it after calling clear_snapshot,
1109   // because clear_snapshot does a sanity check which includes checking the
1110   // XTree.
1111   SXPt* tmp_sxpt = snapshot->alloc_sxpt;
1112   clear_snapshot(snapshot, /*do_sanity_check*/True);
1113   if (tmp_sxpt) {
1114      free_SXTree(tmp_sxpt);
1115   }
1116}
1117
1118static void VERB_snapshot(Int verbosity, const HChar* prefix, Int i)
1119{
1120   Snapshot* snapshot = &snapshots[i];
1121   const HChar* suffix;
1122   switch (snapshot->kind) {
1123   case Peak:   suffix = "p";                                            break;
1124   case Normal: suffix = ( is_detailed_snapshot(snapshot) ? "d" : "." ); break;
1125   case Unused: suffix = "u";                                            break;
1126   default:
1127      tl_assert2(0, "VERB_snapshot: unknown snapshot kind: %d", snapshot->kind);
1128   }
1129   VERB(verbosity, "%s S%s%3d (t:%lld, hp:%ld, ex:%ld, st:%ld)\n",
1130      prefix, suffix, i,
1131      snapshot->time,
1132      snapshot->heap_szB,
1133      snapshot->heap_extra_szB,
1134      snapshot->stacks_szB
1135   );
1136}
1137
1138// Cull half the snapshots;  we choose those that represent the smallest
1139// time-spans, because that gives us the most even distribution of snapshots
1140// over time.  (It's possible to lose interesting spikes, however.)
1141//
1142// Algorithm for N snapshots:  We find the snapshot representing the smallest
1143// timeframe, and remove it.  We repeat this until (N/2) snapshots are gone.
1144// We have to do this one snapshot at a time, rather than finding the (N/2)
1145// smallest snapshots in one hit, because when a snapshot is removed, its
1146// neighbours immediately cover greater timespans.  So it's O(N^2), but N is
1147// small, and it's not done very often.
1148//
1149// Once we're done, we return the new smallest interval between snapshots.
1150// That becomes our minimum time interval.
1151static UInt cull_snapshots(void)
1152{
1153   Int  i, jp, j, jn, min_timespan_i;
1154   Int  n_deleted = 0;
1155   Time min_timespan;
1156
1157   n_cullings++;
1158
1159   // Sets j to the index of the first not-yet-removed snapshot at or after i
1160   #define FIND_SNAPSHOT(i, j) \
1161      for (j = i; \
1162           j < clo_max_snapshots && !is_snapshot_in_use(&snapshots[j]); \
1163           j++) { }
1164
1165   VERB(2, "Culling...\n");
1166
1167   // First we remove enough snapshots by clearing them in-place.  Once
1168   // that's done, we can slide the remaining ones down.
1169   for (i = 0; i < clo_max_snapshots/2; i++) {
1170      // Find the snapshot representing the smallest timespan.  The timespan
1171      // for snapshot n = d(N-1,N)+d(N,N+1), where d(A,B) is the time between
1172      // snapshot A and B.  We don't consider the first and last snapshots for
1173      // removal.
1174      Snapshot* min_snapshot;
1175      Int min_j;
1176
1177      // Initial triple: (prev, curr, next) == (jp, j, jn)
1178      // Initial min_timespan is the first one.
1179      jp = 0;
1180      FIND_SNAPSHOT(1,   j);
1181      FIND_SNAPSHOT(j+1, jn);
1182      min_timespan = 0x7fffffffffffffffLL;
1183      min_j        = -1;
1184      while (jn < clo_max_snapshots) {
1185         Time timespan = snapshots[jn].time - snapshots[jp].time;
1186         tl_assert(timespan >= 0);
1187         // Nb: We never cull the peak snapshot.
1188         if (Peak != snapshots[j].kind && timespan < min_timespan) {
1189            min_timespan = timespan;
1190            min_j        = j;
1191         }
1192         // Move on to next triple
1193         jp = j;
1194         j  = jn;
1195         FIND_SNAPSHOT(jn+1, jn);
1196      }
1197      // We've found the least important snapshot, now delete it.  First
1198      // print it if necessary.
1199      tl_assert(-1 != min_j);    // Check we found a minimum.
1200      min_snapshot = & snapshots[ min_j ];
1201      if (VG_(clo_verbosity) > 1) {
1202         HChar buf[64];
1203         VG_(snprintf)(buf, 64, " %3d (t-span = %lld)", i, min_timespan);
1204         VERB_snapshot(2, buf, min_j);
1205      }
1206      delete_snapshot(min_snapshot);
1207      n_deleted++;
1208   }
1209
1210   // Slide down the remaining snapshots over the removed ones.  First set i
1211   // to point to the first empty slot, and j to the first full slot after
1212   // i.  Then slide everything down.
1213   for (i = 0;  is_snapshot_in_use( &snapshots[i] ); i++) { }
1214   for (j = i; !is_snapshot_in_use( &snapshots[j] ); j++) { }
1215   for (  ; j < clo_max_snapshots; j++) {
1216      if (is_snapshot_in_use( &snapshots[j] )) {
1217         snapshots[i++] = snapshots[j];
1218         clear_snapshot(&snapshots[j], /*do_sanity_check*/True);
1219      }
1220   }
1221   next_snapshot_i = i;
1222
1223   // Check snapshots array looks ok after changes.
1224   sanity_check_snapshots_array();
1225
1226   // Find the minimum timespan remaining;  that will be our new minimum
1227   // time interval.  Note that above we were finding timespans by measuring
1228   // two intervals around a snapshot that was under consideration for
1229   // deletion.  Here we only measure single intervals because all the
1230   // deletions have occurred.
1231   //
1232   // But we have to be careful -- some snapshots (eg. snapshot 0, and the
1233   // peak snapshot) are uncullable.  If two uncullable snapshots end up
1234   // next to each other, they'll never be culled (assuming the peak doesn't
1235   // change), and the time gap between them will not change.  However, the
1236   // time between the remaining cullable snapshots will grow ever larger.
1237   // This means that the min_timespan found will always be that between the
1238   // two uncullable snapshots, and it will be much smaller than it should
1239   // be.  To avoid this problem, when computing the minimum timespan, we
1240   // ignore any timespans between two uncullable snapshots.
1241   tl_assert(next_snapshot_i > 1);
1242   min_timespan = 0x7fffffffffffffffLL;
1243   min_timespan_i = -1;
1244   for (i = 1; i < next_snapshot_i; i++) {
1245      if (is_uncullable_snapshot(&snapshots[i]) &&
1246          is_uncullable_snapshot(&snapshots[i-1]))
1247      {
1248         VERB(2, "(Ignoring interval %d--%d when computing minimum)\n", i-1, i);
1249      } else {
1250         Time timespan = snapshots[i].time - snapshots[i-1].time;
1251         tl_assert(timespan >= 0);
1252         if (timespan < min_timespan) {
1253            min_timespan = timespan;
1254            min_timespan_i = i;
1255         }
1256      }
1257   }
1258   tl_assert(-1 != min_timespan_i);    // Check we found a minimum.
1259
1260   // Print remaining snapshots, if necessary.
1261   if (VG_(clo_verbosity) > 1) {
1262      VERB(2, "Finished culling (%3d of %3d deleted)\n",
1263         n_deleted, clo_max_snapshots);
1264      for (i = 0; i < next_snapshot_i; i++) {
1265         VERB_snapshot(2, "  post-cull", i);
1266      }
1267      VERB(2, "New time interval = %lld (between snapshots %d and %d)\n",
1268         min_timespan, min_timespan_i-1, min_timespan_i);
1269   }
1270
1271   return min_timespan;
1272}
1273
1274static Time get_time(void)
1275{
1276   // Get current time, in whatever time unit we're using.
1277   if (clo_time_unit == TimeI) {
1278      return guest_instrs_executed;
1279   } else if (clo_time_unit == TimeMS) {
1280      // Some stuff happens between the millisecond timer being initialised
1281      // to zero and us taking our first snapshot.  We determine that time
1282      // gap so we can subtract it from all subsequent times so that our
1283      // first snapshot is considered to be at t = 0ms.  Unfortunately, a
1284      // bunch of symbols get read after the first snapshot is taken but
1285      // before the second one (which is triggered by the first allocation),
1286      // so when the time-unit is 'ms' we always have a big gap between the
1287      // first two snapshots.  But at least users won't have to wonder why
1288      // the first snapshot isn't at t=0.
1289      static Bool is_first_get_time = True;
1290      static Time start_time_ms;
1291      if (is_first_get_time) {
1292         start_time_ms = VG_(read_millisecond_timer)();
1293         is_first_get_time = False;
1294         return 0;
1295      } else {
1296         return VG_(read_millisecond_timer)() - start_time_ms;
1297      }
1298   } else if (clo_time_unit == TimeB) {
1299      return total_allocs_deallocs_szB;
1300   } else {
1301      tl_assert2(0, "bad --time-unit value");
1302   }
1303}
1304
1305// Take a snapshot, and only that -- decisions on whether to take a
1306// snapshot, or what kind of snapshot, are made elsewhere.
1307// Nb: we call the arg "my_time" because "time" shadows a global declaration
1308// in /usr/include/time.h on Darwin.
1309static void
1310take_snapshot(Snapshot* snapshot, SnapshotKind kind, Time my_time,
1311              Bool is_detailed)
1312{
1313   tl_assert(!is_snapshot_in_use(snapshot));
1314   if (!clo_pages_as_heap) {
1315      tl_assert(have_started_executing_code);
1316   }
1317
1318   // Heap and heap admin.
1319   if (clo_heap) {
1320      snapshot->heap_szB = heap_szB;
1321      if (is_detailed) {
1322         SizeT total_szB = heap_szB + heap_extra_szB + stacks_szB;
1323         snapshot->alloc_sxpt = dup_XTree(alloc_xpt, total_szB);
1324         tl_assert(           alloc_xpt->szB == heap_szB);
1325         tl_assert(snapshot->alloc_sxpt->szB == heap_szB);
1326      }
1327      snapshot->heap_extra_szB = heap_extra_szB;
1328   }
1329
1330   // Stack(s).
1331   if (clo_stacks) {
1332      snapshot->stacks_szB = stacks_szB;
1333   }
1334
1335   // Rest of snapshot.
1336   snapshot->kind = kind;
1337   snapshot->time = my_time;
1338   sanity_check_snapshot(snapshot);
1339
1340   // Update stats.
1341   if (Peak == kind) n_peak_snapshots++;
1342   if (is_detailed)  n_detailed_snapshots++;
1343   n_real_snapshots++;
1344}
1345
1346
1347// Take a snapshot, if it's time, or if we've hit a peak.
1348static void
1349maybe_take_snapshot(SnapshotKind kind, const HChar* what)
1350{
1351   // 'min_time_interval' is the minimum time interval between snapshots.
1352   // If we try to take a snapshot and less than this much time has passed,
1353   // we don't take it.  It gets larger as the program runs longer.  It's
1354   // initialised to zero so that we begin by taking snapshots as quickly as
1355   // possible.
1356   static Time min_time_interval = 0;
1357   // Zero allows startup snapshot.
1358   static Time earliest_possible_time_of_next_snapshot = 0;
1359   static Int  n_snapshots_since_last_detailed         = 0;
1360   static Int  n_skipped_snapshots_since_last_snapshot = 0;
1361
1362   Snapshot* snapshot;
1363   Bool      is_detailed;
1364   // Nb: we call this variable "my_time" because "time" shadows a global
1365   // declaration in /usr/include/time.h on Darwin.
1366   Time      my_time = get_time();
1367
1368   switch (kind) {
1369    case Normal:
1370      // Only do a snapshot if it's time.
1371      if (my_time < earliest_possible_time_of_next_snapshot) {
1372         n_skipped_snapshots++;
1373         n_skipped_snapshots_since_last_snapshot++;
1374         return;
1375      }
1376      is_detailed = (clo_detailed_freq-1 == n_snapshots_since_last_detailed);
1377      break;
1378
1379    case Peak: {
1380      // Because we're about to do a deallocation, we're coming down from a
1381      // local peak.  If it is (a) actually a global peak, and (b) a certain
1382      // amount bigger than the previous peak, then we take a peak snapshot.
1383      // By not taking a snapshot for every peak, we save a lot of effort --
1384      // because many peaks remain peak only for a short time.
1385      SizeT total_szB = heap_szB + heap_extra_szB + stacks_szB;
1386      SizeT excess_szB_for_new_peak =
1387         (SizeT)((peak_snapshot_total_szB * clo_peak_inaccuracy) / 100);
1388      if (total_szB <= peak_snapshot_total_szB + excess_szB_for_new_peak) {
1389         return;
1390      }
1391      is_detailed = True;
1392      break;
1393    }
1394
1395    default:
1396      tl_assert2(0, "maybe_take_snapshot: unrecognised snapshot kind");
1397   }
1398
1399   // Take the snapshot.
1400   snapshot = & snapshots[next_snapshot_i];
1401   take_snapshot(snapshot, kind, my_time, is_detailed);
1402
1403   // Record if it was detailed.
1404   if (is_detailed) {
1405      n_snapshots_since_last_detailed = 0;
1406   } else {
1407      n_snapshots_since_last_detailed++;
1408   }
1409
1410   // Update peak data, if it's a Peak snapshot.
1411   if (Peak == kind) {
1412      Int i, number_of_peaks_snapshots_found = 0;
1413
1414      // Sanity check the size, then update our recorded peak.
1415      SizeT snapshot_total_szB =
1416         snapshot->heap_szB + snapshot->heap_extra_szB + snapshot->stacks_szB;
1417      tl_assert2(snapshot_total_szB > peak_snapshot_total_szB,
1418         "%ld, %ld\n", snapshot_total_szB, peak_snapshot_total_szB);
1419      peak_snapshot_total_szB = snapshot_total_szB;
1420
1421      // Find the old peak snapshot, if it exists, and mark it as normal.
1422      for (i = 0; i < next_snapshot_i; i++) {
1423         if (Peak == snapshots[i].kind) {
1424            snapshots[i].kind = Normal;
1425            number_of_peaks_snapshots_found++;
1426         }
1427      }
1428      tl_assert(number_of_peaks_snapshots_found <= 1);
1429   }
1430
1431   // Finish up verbosity and stats stuff.
1432   if (n_skipped_snapshots_since_last_snapshot > 0) {
1433      VERB(2, "  (skipped %d snapshot%s)\n",
1434         n_skipped_snapshots_since_last_snapshot,
1435         ( 1 == n_skipped_snapshots_since_last_snapshot ? "" : "s") );
1436   }
1437   VERB_snapshot(2, what, next_snapshot_i);
1438   n_skipped_snapshots_since_last_snapshot = 0;
1439
1440   // Cull the entries, if our snapshot table is full.
1441   next_snapshot_i++;
1442   if (clo_max_snapshots == next_snapshot_i) {
1443      min_time_interval = cull_snapshots();
1444   }
1445
1446   // Work out the earliest time when the next snapshot can happen.
1447   earliest_possible_time_of_next_snapshot = my_time + min_time_interval;
1448}
1449
1450
1451//------------------------------------------------------------//
1452//--- Sanity checking                                      ---//
1453//------------------------------------------------------------//
1454
1455static Bool ms_cheap_sanity_check ( void )
1456{
1457   return True;   // Nothing useful we can cheaply check.
1458}
1459
1460static Bool ms_expensive_sanity_check ( void )
1461{
1462   sanity_check_XTree(alloc_xpt, /*parent*/NULL);
1463   sanity_check_snapshots_array();
1464   return True;
1465}
1466
1467
1468//------------------------------------------------------------//
1469//--- Heap management                                      ---//
1470//------------------------------------------------------------//
1471
1472// Metadata for heap blocks.  Each one contains a pointer to a bottom-XPt,
1473// which is a foothold into the XCon at which it was allocated.  From
1474// HP_Chunks, XPt 'space' fields are incremented (at allocation) and
1475// decremented (at deallocation).
1476//
1477// Nb: first two fields must match core's VgHashNode.
1478typedef
1479   struct _HP_Chunk {
1480      struct _HP_Chunk* next;
1481      Addr              data;       // Ptr to actual block
1482      SizeT             req_szB;    // Size requested
1483      SizeT             slop_szB;   // Extra bytes given above those requested
1484      XPt*              where;      // Where allocated; bottom-XPt
1485   }
1486   HP_Chunk;
1487
1488static VgHashTable malloc_list  = NULL;   // HP_Chunks
1489
1490static void update_alloc_stats(SSizeT szB_delta)
1491{
1492   // Update total_allocs_deallocs_szB.
1493   if (szB_delta < 0) szB_delta = -szB_delta;
1494   total_allocs_deallocs_szB += szB_delta;
1495}
1496
1497static void update_heap_stats(SSizeT heap_szB_delta, Int heap_extra_szB_delta)
1498{
1499   if (heap_szB_delta < 0)
1500      tl_assert(heap_szB >= -heap_szB_delta);
1501   if (heap_extra_szB_delta < 0)
1502      tl_assert(heap_extra_szB >= -heap_extra_szB_delta);
1503
1504   heap_extra_szB += heap_extra_szB_delta;
1505   heap_szB       += heap_szB_delta;
1506
1507   update_alloc_stats(heap_szB_delta + heap_extra_szB_delta);
1508}
1509
1510static
1511void* record_block( ThreadId tid, void* p, SizeT req_szB, SizeT slop_szB,
1512                    Bool exclude_first_entry, Bool maybe_snapshot )
1513{
1514   // Make new HP_Chunk node, add to malloc_list
1515   HP_Chunk* hc = VG_(malloc)("ms.main.rb.1", sizeof(HP_Chunk));
1516   hc->req_szB  = req_szB;
1517   hc->slop_szB = slop_szB;
1518   hc->data     = (Addr)p;
1519   hc->where    = NULL;
1520   VG_(HT_add_node)(malloc_list, hc);
1521
1522   if (clo_heap) {
1523      VERB(3, "<<< record_block (%lu, %lu)\n", req_szB, slop_szB);
1524
1525      hc->where = get_XCon( tid, exclude_first_entry );
1526
1527      if (hc->where) {
1528         // Update statistics.
1529         n_heap_allocs++;
1530
1531         // Update heap stats.
1532         update_heap_stats(req_szB, clo_heap_admin + slop_szB);
1533
1534         // Update XTree.
1535         update_XCon(hc->where, req_szB);
1536
1537         // Maybe take a snapshot.
1538         if (maybe_snapshot) {
1539            maybe_take_snapshot(Normal, "  alloc");
1540         }
1541
1542      } else {
1543         // Ignored allocation.
1544         n_ignored_heap_allocs++;
1545
1546         VERB(3, "(ignored)\n");
1547      }
1548
1549      VERB(3, ">>>\n");
1550   }
1551
1552   return p;
1553}
1554
1555static __inline__
1556void* alloc_and_record_block ( ThreadId tid, SizeT req_szB, SizeT req_alignB,
1557                               Bool is_zeroed )
1558{
1559   SizeT actual_szB, slop_szB;
1560   void* p;
1561
1562   if ((SSizeT)req_szB < 0) return NULL;
1563
1564   // Allocate and zero if necessary.
1565   p = VG_(cli_malloc)( req_alignB, req_szB );
1566   if (!p) {
1567      return NULL;
1568   }
1569   if (is_zeroed) VG_(memset)(p, 0, req_szB);
1570   actual_szB = VG_(malloc_usable_size)(p);
1571   tl_assert(actual_szB >= req_szB);
1572   slop_szB = actual_szB - req_szB;
1573
1574   // Record block.
1575   record_block(tid, p, req_szB, slop_szB, /*exclude_first_entry*/True,
1576                /*maybe_snapshot*/True);
1577
1578   return p;
1579}
1580
1581static __inline__
1582void unrecord_block ( void* p, Bool maybe_snapshot )
1583{
1584   // Remove HP_Chunk from malloc_list
1585   HP_Chunk* hc = VG_(HT_remove)(malloc_list, (UWord)p);
1586   if (NULL == hc) {
1587      return;   // must have been a bogus free()
1588   }
1589
1590   if (clo_heap) {
1591      VERB(3, "<<< unrecord_block\n");
1592
1593      if (hc->where) {
1594         // Update statistics.
1595         n_heap_frees++;
1596
1597         // Maybe take a peak snapshot, since it's a deallocation.
1598         if (maybe_snapshot) {
1599            maybe_take_snapshot(Peak, "de-PEAK");
1600         }
1601
1602         // Update heap stats.
1603         update_heap_stats(-hc->req_szB, -clo_heap_admin - hc->slop_szB);
1604
1605         // Update XTree.
1606         update_XCon(hc->where, -hc->req_szB);
1607
1608         // Maybe take a snapshot.
1609         if (maybe_snapshot) {
1610            maybe_take_snapshot(Normal, "dealloc");
1611         }
1612
1613      } else {
1614         n_ignored_heap_frees++;
1615
1616         VERB(3, "(ignored)\n");
1617      }
1618
1619      VERB(3, ">>> (-%lu, -%lu)\n", hc->req_szB, hc->slop_szB);
1620   }
1621
1622   // Actually free the chunk, and the heap block (if necessary)
1623   VG_(free)( hc );  hc = NULL;
1624}
1625
1626// Nb: --ignore-fn is tricky for realloc.  If the block's original alloc was
1627// ignored, but the realloc is not requested to be ignored, and we are
1628// shrinking the block, then we have to ignore the realloc -- otherwise we
1629// could end up with negative heap sizes.  This isn't a danger if we are
1630// growing such a block, but for consistency (it also simplifies things) we
1631// ignore such reallocs as well.
1632static __inline__
1633void* realloc_block ( ThreadId tid, void* p_old, SizeT new_req_szB )
1634{
1635   HP_Chunk* hc;
1636   void*     p_new;
1637   SizeT     old_req_szB, old_slop_szB, new_slop_szB, new_actual_szB;
1638   XPt      *old_where, *new_where;
1639   Bool      is_ignored = False;
1640
1641   // Remove the old block
1642   hc = VG_(HT_remove)(malloc_list, (UWord)p_old);
1643   if (hc == NULL) {
1644      return NULL;   // must have been a bogus realloc()
1645   }
1646
1647   old_req_szB  = hc->req_szB;
1648   old_slop_szB = hc->slop_szB;
1649
1650   tl_assert(!clo_pages_as_heap);  // Shouldn't be here if --pages-as-heap=yes.
1651   if (clo_heap) {
1652      VERB(3, "<<< realloc_block (%lu)\n", new_req_szB);
1653
1654      if (hc->where) {
1655         // Update statistics.
1656         n_heap_reallocs++;
1657
1658         // Maybe take a peak snapshot, if it's (effectively) a deallocation.
1659         if (new_req_szB < old_req_szB) {
1660            maybe_take_snapshot(Peak, "re-PEAK");
1661         }
1662      } else {
1663         // The original malloc was ignored, so we have to ignore the
1664         // realloc as well.
1665         is_ignored = True;
1666      }
1667   }
1668
1669   // Actually do the allocation, if necessary.
1670   if (new_req_szB <= old_req_szB + old_slop_szB) {
1671      // New size is smaller or same;  block not moved.
1672      p_new = p_old;
1673      new_slop_szB = old_slop_szB + (old_req_szB - new_req_szB);
1674
1675   } else {
1676      // New size is bigger;  make new block, copy shared contents, free old.
1677      p_new = VG_(cli_malloc)(VG_(clo_alignment), new_req_szB);
1678      if (!p_new) {
1679         // Nb: if realloc fails, NULL is returned but the old block is not
1680         // touched.  What an awful function.
1681         return NULL;
1682      }
1683      VG_(memcpy)(p_new, p_old, old_req_szB + old_slop_szB);
1684      VG_(cli_free)(p_old);
1685      new_actual_szB = VG_(malloc_usable_size)(p_new);
1686      tl_assert(new_actual_szB >= new_req_szB);
1687      new_slop_szB = new_actual_szB - new_req_szB;
1688   }
1689
1690   if (p_new) {
1691      // Update HP_Chunk.
1692      hc->data     = (Addr)p_new;
1693      hc->req_szB  = new_req_szB;
1694      hc->slop_szB = new_slop_szB;
1695      old_where    = hc->where;
1696      hc->where    = NULL;
1697
1698      // Update XTree.
1699      if (clo_heap) {
1700         new_where = get_XCon( tid, /*exclude_first_entry*/True);
1701         if (!is_ignored && new_where) {
1702            hc->where = new_where;
1703            update_XCon(old_where, -old_req_szB);
1704            update_XCon(new_where,  new_req_szB);
1705         } else {
1706            // The realloc itself is ignored.
1707            is_ignored = True;
1708
1709            // Update statistics.
1710            n_ignored_heap_reallocs++;
1711         }
1712      }
1713   }
1714
1715   // Now insert the new hc (with a possibly new 'data' field) into
1716   // malloc_list.  If this realloc() did not increase the memory size, we
1717   // will have removed and then re-added hc unnecessarily.  But that's ok
1718   // because shrinking a block with realloc() is (presumably) much rarer
1719   // than growing it, and this way simplifies the growing case.
1720   VG_(HT_add_node)(malloc_list, hc);
1721
1722   if (clo_heap) {
1723      if (!is_ignored) {
1724         // Update heap stats.
1725         update_heap_stats(new_req_szB - old_req_szB,
1726                          new_slop_szB - old_slop_szB);
1727
1728         // Maybe take a snapshot.
1729         maybe_take_snapshot(Normal, "realloc");
1730      } else {
1731
1732         VERB(3, "(ignored)\n");
1733      }
1734
1735      VERB(3, ">>> (%ld, %ld)\n",
1736         new_req_szB - old_req_szB, new_slop_szB - old_slop_szB);
1737   }
1738
1739   return p_new;
1740}
1741
1742
1743//------------------------------------------------------------//
1744//--- malloc() et al replacement wrappers                  ---//
1745//------------------------------------------------------------//
1746
1747static void* ms_malloc ( ThreadId tid, SizeT szB )
1748{
1749   return alloc_and_record_block( tid, szB, VG_(clo_alignment), /*is_zeroed*/False );
1750}
1751
1752static void* ms___builtin_new ( ThreadId tid, SizeT szB )
1753{
1754   return alloc_and_record_block( tid, szB, VG_(clo_alignment), /*is_zeroed*/False );
1755}
1756
1757static void* ms___builtin_vec_new ( ThreadId tid, SizeT szB )
1758{
1759   return alloc_and_record_block( tid, szB, VG_(clo_alignment), /*is_zeroed*/False );
1760}
1761
1762static void* ms_calloc ( ThreadId tid, SizeT m, SizeT szB )
1763{
1764   return alloc_and_record_block( tid, m*szB, VG_(clo_alignment), /*is_zeroed*/True );
1765}
1766
1767static void *ms_memalign ( ThreadId tid, SizeT alignB, SizeT szB )
1768{
1769   return alloc_and_record_block( tid, szB, alignB, False );
1770}
1771
1772static void ms_free ( ThreadId tid __attribute__((unused)), void* p )
1773{
1774   unrecord_block(p, /*maybe_snapshot*/True);
1775   VG_(cli_free)(p);
1776}
1777
1778static void ms___builtin_delete ( ThreadId tid, void* p )
1779{
1780   unrecord_block(p, /*maybe_snapshot*/True);
1781   VG_(cli_free)(p);
1782}
1783
1784static void ms___builtin_vec_delete ( ThreadId tid, void* p )
1785{
1786   unrecord_block(p, /*maybe_snapshot*/True);
1787   VG_(cli_free)(p);
1788}
1789
1790static void* ms_realloc ( ThreadId tid, void* p_old, SizeT new_szB )
1791{
1792   return realloc_block(tid, p_old, new_szB);
1793}
1794
1795static SizeT ms_malloc_usable_size ( ThreadId tid, void* p )
1796{
1797   HP_Chunk* hc = VG_(HT_lookup)( malloc_list, (UWord)p );
1798
1799   return ( hc ? hc->req_szB + hc->slop_szB : 0 );
1800}
1801
1802//------------------------------------------------------------//
1803//--- Page handling                                        ---//
1804//------------------------------------------------------------//
1805
1806static
1807void ms_record_page_mem ( Addr a, SizeT len )
1808{
1809   ThreadId tid = VG_(get_running_tid)();
1810   Addr end;
1811   tl_assert(VG_IS_PAGE_ALIGNED(len));
1812   tl_assert(len >= VKI_PAGE_SIZE);
1813   // Record the first N-1 pages as blocks, but don't do any snapshots.
1814   for (end = a + len - VKI_PAGE_SIZE; a < end; a += VKI_PAGE_SIZE) {
1815      record_block( tid, (void*)a, VKI_PAGE_SIZE, /*slop_szB*/0,
1816                    /*exclude_first_entry*/False, /*maybe_snapshot*/False );
1817   }
1818   // Record the last page as a block, and maybe do a snapshot afterwards.
1819   record_block( tid, (void*)a, VKI_PAGE_SIZE, /*slop_szB*/0,
1820                 /*exclude_first_entry*/False, /*maybe_snapshot*/True );
1821}
1822
1823static
1824void ms_unrecord_page_mem( Addr a, SizeT len )
1825{
1826   Addr end;
1827   tl_assert(VG_IS_PAGE_ALIGNED(len));
1828   tl_assert(len >= VKI_PAGE_SIZE);
1829   for (end = a + len - VKI_PAGE_SIZE; a < end; a += VKI_PAGE_SIZE) {
1830      unrecord_block((void*)a, /*maybe_snapshot*/False);
1831   }
1832   unrecord_block((void*)a, /*maybe_snapshot*/True);
1833}
1834
1835//------------------------------------------------------------//
1836
1837static
1838void ms_new_mem_mmap ( Addr a, SizeT len,
1839                       Bool rr, Bool ww, Bool xx, ULong di_handle )
1840{
1841   tl_assert(VG_IS_PAGE_ALIGNED(len));
1842   ms_record_page_mem(a, len);
1843}
1844
1845static
1846void ms_new_mem_startup( Addr a, SizeT len,
1847                         Bool rr, Bool ww, Bool xx, ULong di_handle )
1848{
1849   // startup maps are always be page-sized, except the trampoline page is
1850   // marked by the core as only being the size of the trampoline itself,
1851   // which is something like 57 bytes.  Round it up to page size.
1852   len = VG_PGROUNDUP(len);
1853   ms_record_page_mem(a, len);
1854}
1855
1856static
1857void ms_new_mem_brk ( Addr a, SizeT len, ThreadId tid )
1858{
1859   // brk limit is not necessarily aligned on a page boundary.
1860   // If new memory being brk-ed implies to allocate a new page,
1861   // then call ms_record_page_mem with page aligned parameters
1862   // otherwise just ignore.
1863   Addr old_bottom_page = VG_PGROUNDDN(a - 1);
1864   Addr new_top_page = VG_PGROUNDDN(a + len - 1);
1865   if (old_bottom_page != new_top_page)
1866      ms_record_page_mem(VG_PGROUNDDN(a),
1867                         (new_top_page - old_bottom_page));
1868}
1869
1870static
1871void ms_copy_mem_remap( Addr from, Addr to, SizeT len)
1872{
1873   tl_assert(VG_IS_PAGE_ALIGNED(len));
1874   ms_unrecord_page_mem(from, len);
1875   ms_record_page_mem(to, len);
1876}
1877
1878static
1879void ms_die_mem_munmap( Addr a, SizeT len )
1880{
1881   tl_assert(VG_IS_PAGE_ALIGNED(len));
1882   ms_unrecord_page_mem(a, len);
1883}
1884
1885static
1886void ms_die_mem_brk( Addr a, SizeT len )
1887{
1888   // Call ms_unrecord_page_mem only if one or more pages are de-allocated.
1889   // See ms_new_mem_brk for more details.
1890   Addr new_bottom_page = VG_PGROUNDDN(a - 1);
1891   Addr old_top_page = VG_PGROUNDDN(a + len - 1);
1892   if (old_top_page != new_bottom_page)
1893      ms_unrecord_page_mem(VG_PGROUNDDN(a),
1894                           (old_top_page - new_bottom_page));
1895
1896}
1897
1898//------------------------------------------------------------//
1899//--- Stacks                                               ---//
1900//------------------------------------------------------------//
1901
1902// We really want the inlining to occur...
1903#define INLINE    inline __attribute__((always_inline))
1904
1905static void update_stack_stats(SSizeT stack_szB_delta)
1906{
1907   if (stack_szB_delta < 0) tl_assert(stacks_szB >= -stack_szB_delta);
1908   stacks_szB += stack_szB_delta;
1909
1910   update_alloc_stats(stack_szB_delta);
1911}
1912
1913static INLINE void new_mem_stack_2(SizeT len, const HChar* what)
1914{
1915   if (have_started_executing_code) {
1916      VERB(3, "<<< new_mem_stack (%ld)\n", len);
1917      n_stack_allocs++;
1918      update_stack_stats(len);
1919      maybe_take_snapshot(Normal, what);
1920      VERB(3, ">>>\n");
1921   }
1922}
1923
1924static INLINE void die_mem_stack_2(SizeT len, const HChar* what)
1925{
1926   if (have_started_executing_code) {
1927      VERB(3, "<<< die_mem_stack (%ld)\n", -len);
1928      n_stack_frees++;
1929      maybe_take_snapshot(Peak,   "stkPEAK");
1930      update_stack_stats(-len);
1931      maybe_take_snapshot(Normal, what);
1932      VERB(3, ">>>\n");
1933   }
1934}
1935
1936static void new_mem_stack(Addr a, SizeT len)
1937{
1938   new_mem_stack_2(len, "stk-new");
1939}
1940
1941static void die_mem_stack(Addr a, SizeT len)
1942{
1943   die_mem_stack_2(len, "stk-die");
1944}
1945
1946static void new_mem_stack_signal(Addr a, SizeT len, ThreadId tid)
1947{
1948   new_mem_stack_2(len, "sig-new");
1949}
1950
1951static void die_mem_stack_signal(Addr a, SizeT len)
1952{
1953   die_mem_stack_2(len, "sig-die");
1954}
1955
1956
1957//------------------------------------------------------------//
1958//--- Client Requests                                      ---//
1959//------------------------------------------------------------//
1960
1961static void print_monitor_help ( void )
1962{
1963   VG_(gdb_printf) ("\n");
1964   VG_(gdb_printf) ("massif monitor commands:\n");
1965   VG_(gdb_printf) ("  snapshot [<filename>]\n");
1966   VG_(gdb_printf) ("  detailed_snapshot [<filename>]\n");
1967   VG_(gdb_printf) ("       takes a snapshot (or a detailed snapshot)\n");
1968   VG_(gdb_printf) ("       and saves it in <filename>\n");
1969   VG_(gdb_printf) ("             default <filename> is massif.vgdb.out\n");
1970   VG_(gdb_printf) ("\n");
1971}
1972
1973
1974/* Forward declaration.
1975   return True if request recognised, False otherwise */
1976static Bool handle_gdb_monitor_command (ThreadId tid, HChar *req);
1977static Bool ms_handle_client_request ( ThreadId tid, UWord* argv, UWord* ret )
1978{
1979   switch (argv[0]) {
1980   case VG_USERREQ__MALLOCLIKE_BLOCK: {
1981      void* p   = (void*)argv[1];
1982      SizeT szB =        argv[2];
1983      record_block( tid, p, szB, /*slop_szB*/0, /*exclude_first_entry*/False,
1984                    /*maybe_snapshot*/True );
1985      *ret = 0;
1986      return True;
1987   }
1988   case VG_USERREQ__RESIZEINPLACE_BLOCK: {
1989      void* p        = (void*)argv[1];
1990      SizeT newSizeB =       argv[3];
1991
1992      unrecord_block(p, /*maybe_snapshot*/True);
1993      record_block(tid, p, newSizeB, /*slop_szB*/0,
1994                   /*exclude_first_entry*/False, /*maybe_snapshot*/True);
1995      return True;
1996   }
1997   case VG_USERREQ__FREELIKE_BLOCK: {
1998      void* p = (void*)argv[1];
1999      unrecord_block(p, /*maybe_snapshot*/True);
2000      *ret = 0;
2001      return True;
2002   }
2003   case VG_USERREQ__GDB_MONITOR_COMMAND: {
2004     Bool handled = handle_gdb_monitor_command (tid, (HChar*)argv[1]);
2005     if (handled)
2006       *ret = 1;
2007     else
2008       *ret = 0;
2009     return handled;
2010   }
2011
2012   default:
2013      *ret = 0;
2014      return False;
2015   }
2016}
2017
2018//------------------------------------------------------------//
2019//--- Instrumentation                                      ---//
2020//------------------------------------------------------------//
2021
2022static void add_counter_update(IRSB* sbOut, Int n)
2023{
2024   #if defined(VG_BIGENDIAN)
2025   # define END Iend_BE
2026   #elif defined(VG_LITTLEENDIAN)
2027   # define END Iend_LE
2028   #else
2029   # error "Unknown endianness"
2030   #endif
2031   // Add code to increment 'guest_instrs_executed' by 'n', like this:
2032   //   WrTmp(t1, Load64(&guest_instrs_executed))
2033   //   WrTmp(t2, Add64(RdTmp(t1), Const(n)))
2034   //   Store(&guest_instrs_executed, t2)
2035   IRTemp t1 = newIRTemp(sbOut->tyenv, Ity_I64);
2036   IRTemp t2 = newIRTemp(sbOut->tyenv, Ity_I64);
2037   IRExpr* counter_addr = mkIRExpr_HWord( (HWord)&guest_instrs_executed );
2038
2039   IRStmt* st1 = IRStmt_WrTmp(t1, IRExpr_Load(END, Ity_I64, counter_addr));
2040   IRStmt* st2 =
2041      IRStmt_WrTmp(t2,
2042                   IRExpr_Binop(Iop_Add64, IRExpr_RdTmp(t1),
2043                                           IRExpr_Const(IRConst_U64(n))));
2044   IRStmt* st3 = IRStmt_Store(END, counter_addr, IRExpr_RdTmp(t2));
2045
2046   addStmtToIRSB( sbOut, st1 );
2047   addStmtToIRSB( sbOut, st2 );
2048   addStmtToIRSB( sbOut, st3 );
2049}
2050
2051static IRSB* ms_instrument2( IRSB* sbIn )
2052{
2053   Int   i, n = 0;
2054   IRSB* sbOut;
2055
2056   // We increment the instruction count in two places:
2057   // - just before any Ist_Exit statements;
2058   // - just before the IRSB's end.
2059   // In the former case, we zero 'n' and then continue instrumenting.
2060
2061   sbOut = deepCopyIRSBExceptStmts(sbIn);
2062
2063   for (i = 0; i < sbIn->stmts_used; i++) {
2064      IRStmt* st = sbIn->stmts[i];
2065
2066      if (!st || st->tag == Ist_NoOp) continue;
2067
2068      if (st->tag == Ist_IMark) {
2069         n++;
2070      } else if (st->tag == Ist_Exit) {
2071         if (n > 0) {
2072            // Add an increment before the Exit statement, then reset 'n'.
2073            add_counter_update(sbOut, n);
2074            n = 0;
2075         }
2076      }
2077      addStmtToIRSB( sbOut, st );
2078   }
2079
2080   if (n > 0) {
2081      // Add an increment before the SB end.
2082      add_counter_update(sbOut, n);
2083   }
2084   return sbOut;
2085}
2086
2087static
2088IRSB* ms_instrument ( VgCallbackClosure* closure,
2089                      IRSB* sbIn,
2090                      VexGuestLayout* layout,
2091                      VexGuestExtents* vge,
2092                      VexArchInfo* archinfo_host,
2093                      IRType gWordTy, IRType hWordTy )
2094{
2095   if (! have_started_executing_code) {
2096      // Do an initial sample to guarantee that we have at least one.
2097      // We use 'maybe_take_snapshot' instead of 'take_snapshot' to ensure
2098      // 'maybe_take_snapshot's internal static variables are initialised.
2099      have_started_executing_code = True;
2100      maybe_take_snapshot(Normal, "startup");
2101   }
2102
2103   if      (clo_time_unit == TimeI)  { return ms_instrument2(sbIn); }
2104   else if (clo_time_unit == TimeMS) { return sbIn; }
2105   else if (clo_time_unit == TimeB)  { return sbIn; }
2106   else                              { tl_assert2(0, "bad --time-unit value"); }
2107}
2108
2109
2110//------------------------------------------------------------//
2111//--- Writing snapshots                                    ---//
2112//------------------------------------------------------------//
2113
2114HChar FP_buf[BUF_LEN];
2115
2116// XXX: implement f{,n}printf in m_libcprint.c eventually, and use it here.
2117// Then change Cachegrind to use it too.
2118#define FP(format, args...) ({ \
2119   VG_(snprintf)(FP_buf, BUF_LEN, format, ##args); \
2120   FP_buf[BUF_LEN-1] = '\0';  /* Make sure the string is terminated. */ \
2121   VG_(write)(fd, (void*)FP_buf, VG_(strlen)(FP_buf)); \
2122})
2123
2124// Nb: uses a static buffer, each call trashes the last string returned.
2125static const HChar* make_perc(double x)
2126{
2127   static HChar mbuf[32];
2128
2129   VG_(percentify)((ULong)(x * 100), 10000, 2, 6, mbuf);
2130   // XXX: this is bogus if the denominator was zero -- resulting string is
2131   // something like "0 --%")
2132   if (' ' == mbuf[0]) mbuf[0] = '0';
2133   return mbuf;
2134}
2135
2136static void pp_snapshot_SXPt(Int fd, SXPt* sxpt, Int depth, HChar* depth_str,
2137                            Int depth_str_len,
2138                            SizeT snapshot_heap_szB, SizeT snapshot_total_szB)
2139{
2140   Int   i, j, n_insig_children_sxpts;
2141   SXPt* child = NULL;
2142
2143   // Used for printing function names.  Is made static to keep it out
2144   // of the stack frame -- this function is recursive.  Obviously this
2145   // now means its contents are trashed across the recursive call.
2146   static HChar ip_desc_array[BUF_LEN];
2147   const HChar* ip_desc = ip_desc_array;
2148
2149   switch (sxpt->tag) {
2150    case SigSXPt:
2151      // Print the SXPt itself.
2152      if (0 == depth) {
2153         if (clo_heap) {
2154            ip_desc =
2155               ( clo_pages_as_heap
2156               ? "(page allocation syscalls) mmap/mremap/brk, --alloc-fns, etc."
2157               : "(heap allocation functions) malloc/new/new[], --alloc-fns, etc."
2158               );
2159         } else {
2160            // XXX: --alloc-fns?
2161
2162            // Nick thinks this case cannot happen. ip_desc_array would be
2163            // conceptually uninitialised here. Therefore:
2164            tl_assert2(0, "pp_snapshot_SXPt: unexpected");
2165         }
2166      } else {
2167         // If it's main-or-below-main, we (if appropriate) ignore everything
2168         // below it by pretending it has no children.
2169         if ( ! VG_(clo_show_below_main) ) {
2170            Vg_FnNameKind kind = VG_(get_fnname_kind_from_IP)(sxpt->Sig.ip);
2171            if (Vg_FnNameMain == kind || Vg_FnNameBelowMain == kind) {
2172               sxpt->Sig.n_children = 0;
2173            }
2174         }
2175
2176         // We need the -1 to get the line number right, But I'm not sure why.
2177         ip_desc = VG_(describe_IP)(sxpt->Sig.ip-1, ip_desc_array, BUF_LEN);
2178      }
2179
2180      // Do the non-ip_desc part first...
2181      FP("%sn%d: %lu ", depth_str, sxpt->Sig.n_children, sxpt->szB);
2182
2183      // For ip_descs beginning with "0xABCD...:" addresses, we first
2184      // measure the length of the "0xabcd: " address at the start of the
2185      // ip_desc.
2186      j = 0;
2187      if ('0' == ip_desc[0] && 'x' == ip_desc[1]) {
2188         j = 2;
2189         while (True) {
2190            if (ip_desc[j]) {
2191               if (':' == ip_desc[j]) break;
2192               j++;
2193            } else {
2194               tl_assert2(0, "ip_desc has unexpected form: %s\n", ip_desc);
2195            }
2196         }
2197      }
2198      // Nb: We treat this specially (ie. we don't use FP) so that if the
2199      // ip_desc is too long (eg. due to a long C++ function name), it'll
2200      // get truncated, but the '\n' is still there so its a valid file.
2201      // (At one point we were truncating without adding the '\n', which
2202      // caused bug #155929.)
2203      //
2204      // Also, we account for the length of the address in ip_desc when
2205      // truncating.  (The longest address we could have is 18 chars:  "0x"
2206      // plus 16 address digits.)  This ensures that the truncated function
2207      // name always has the same length, which makes truncation
2208      // deterministic and thus makes testing easier.
2209      tl_assert(j <= 18);
2210      VG_(snprintf)(FP_buf, BUF_LEN, "%s\n", ip_desc);
2211      FP_buf[BUF_LEN-18+j-5] = '.';    // "..." at the end make the
2212      FP_buf[BUF_LEN-18+j-4] = '.';    //   truncation more obvious.
2213      FP_buf[BUF_LEN-18+j-3] = '.';
2214      FP_buf[BUF_LEN-18+j-2] = '\n';   // The last char is '\n'.
2215      FP_buf[BUF_LEN-18+j-1] = '\0';   // The string is terminated.
2216      VG_(write)(fd, (void*)FP_buf, VG_(strlen)(FP_buf));
2217
2218      // Indent.
2219      tl_assert(depth+1 < depth_str_len-1);    // -1 for end NUL char
2220      depth_str[depth+0] = ' ';
2221      depth_str[depth+1] = '\0';
2222
2223      // Sort SXPt's children by szB (reverse order:  biggest to smallest).
2224      // Nb: we sort them here, rather than earlier (eg. in dup_XTree), for
2225      // two reasons.  First, if we do it during dup_XTree, it can get
2226      // expensive (eg. 15% of execution time for konqueror
2227      // startup/shutdown).  Second, this way we get the Insig SXPt (if one
2228      // is present) in its sorted position, not at the end.
2229      VG_(ssort)(sxpt->Sig.children, sxpt->Sig.n_children, sizeof(SXPt*),
2230                 SXPt_revcmp_szB);
2231
2232      // Print the SXPt's children.  They should already be in sorted order.
2233      n_insig_children_sxpts = 0;
2234      for (i = 0; i < sxpt->Sig.n_children; i++) {
2235         child = sxpt->Sig.children[i];
2236
2237         if (InsigSXPt == child->tag)
2238            n_insig_children_sxpts++;
2239
2240         // Ok, print the child.  NB: contents of ip_desc_array will be
2241         // trashed by this recursive call.  Doesn't matter currently,
2242         // but worth noting.
2243         pp_snapshot_SXPt(fd, child, depth+1, depth_str, depth_str_len,
2244            snapshot_heap_szB, snapshot_total_szB);
2245      }
2246
2247      // Unindent.
2248      depth_str[depth+0] = '\0';
2249      depth_str[depth+1] = '\0';
2250
2251      // There should be 0 or 1 Insig children SXPts.
2252      tl_assert(n_insig_children_sxpts <= 1);
2253      break;
2254
2255    case InsigSXPt: {
2256      const HChar* s = ( 1 == sxpt->Insig.n_xpts ? "," : "s, all" );
2257      FP("%sn0: %lu in %d place%s below massif's threshold (%s)\n",
2258         depth_str, sxpt->szB, sxpt->Insig.n_xpts, s,
2259         make_perc(clo_threshold));
2260      break;
2261    }
2262
2263    default:
2264      tl_assert2(0, "pp_snapshot_SXPt: unrecognised SXPt tag");
2265   }
2266}
2267
2268static void pp_snapshot(Int fd, Snapshot* snapshot, Int snapshot_n)
2269{
2270   sanity_check_snapshot(snapshot);
2271
2272   FP("#-----------\n");
2273   FP("snapshot=%d\n", snapshot_n);
2274   FP("#-----------\n");
2275   FP("time=%lld\n",            snapshot->time);
2276   FP("mem_heap_B=%lu\n",       snapshot->heap_szB);
2277   FP("mem_heap_extra_B=%lu\n", snapshot->heap_extra_szB);
2278   FP("mem_stacks_B=%lu\n",     snapshot->stacks_szB);
2279
2280   if (is_detailed_snapshot(snapshot)) {
2281      // Detailed snapshot -- print heap tree.
2282      Int   depth_str_len = clo_depth + 3;
2283      HChar* depth_str = VG_(malloc)("ms.main.pps.1",
2284                                     sizeof(HChar) * depth_str_len);
2285      SizeT snapshot_total_szB =
2286         snapshot->heap_szB + snapshot->heap_extra_szB + snapshot->stacks_szB;
2287      depth_str[0] = '\0';   // Initialise depth_str to "".
2288
2289      FP("heap_tree=%s\n", ( Peak == snapshot->kind ? "peak" : "detailed" ));
2290      pp_snapshot_SXPt(fd, snapshot->alloc_sxpt, 0, depth_str,
2291                       depth_str_len, snapshot->heap_szB,
2292                       snapshot_total_szB);
2293
2294      VG_(free)(depth_str);
2295
2296   } else {
2297      FP("heap_tree=empty\n");
2298   }
2299}
2300
2301static void write_snapshots_to_file(const HChar* massif_out_file,
2302                                    Snapshot snapshots_array[],
2303                                    Int nr_elements)
2304{
2305   Int i, fd;
2306   SysRes sres;
2307
2308   sres = VG_(open)(massif_out_file, VKI_O_CREAT|VKI_O_TRUNC|VKI_O_WRONLY,
2309                                     VKI_S_IRUSR|VKI_S_IWUSR);
2310   if (sr_isError(sres)) {
2311      // If the file can't be opened for whatever reason (conflict
2312      // between multiple cachegrinded processes?), give up now.
2313      VG_(umsg)("error: can't open output file '%s'\n", massif_out_file );
2314      VG_(umsg)("       ... so profiling results will be missing.\n");
2315      return;
2316   } else {
2317      fd = sr_Res(sres);
2318   }
2319
2320   // Print massif-specific options that were used.
2321   // XXX: is it worth having a "desc:" line?  Could just call it "options:"
2322   // -- this file format isn't as generic as Cachegrind's, so the
2323   // implied genericity of "desc:" is bogus.
2324   FP("desc:");
2325   for (i = 0; i < VG_(sizeXA)(args_for_massif); i++) {
2326      HChar* arg = *(HChar**)VG_(indexXA)(args_for_massif, i);
2327      FP(" %s", arg);
2328   }
2329   if (0 == i) FP(" (none)");
2330   FP("\n");
2331
2332   // Print "cmd:" line.
2333   FP("cmd: ");
2334   if (VG_(args_the_exename)) {
2335      FP("%s", VG_(args_the_exename));
2336      for (i = 0; i < VG_(sizeXA)( VG_(args_for_client) ); i++) {
2337         HChar* arg = * (HChar**) VG_(indexXA)( VG_(args_for_client), i );
2338         if (arg)
2339            FP(" %s", arg);
2340      }
2341   } else {
2342      FP(" ???");
2343   }
2344   FP("\n");
2345
2346   FP("time_unit: %s\n", TimeUnit_to_string(clo_time_unit));
2347
2348   for (i = 0; i < nr_elements; i++) {
2349      Snapshot* snapshot = & snapshots_array[i];
2350      pp_snapshot(fd, snapshot, i);     // Detailed snapshot!
2351   }
2352   VG_(close) (fd);
2353}
2354
2355static void write_snapshots_array_to_file(void)
2356{
2357   // Setup output filename.  Nb: it's important to do this now, ie. as late
2358   // as possible.  If we do it at start-up and the program forks and the
2359   // output file format string contains a %p (pid) specifier, both the
2360   // parent and child will incorrectly write to the same file;  this
2361   // happened in 3.3.0.
2362   HChar* massif_out_file =
2363      VG_(expand_file_name)("--massif-out-file", clo_massif_out_file);
2364   write_snapshots_to_file (massif_out_file, snapshots, next_snapshot_i);
2365   VG_(free)(massif_out_file);
2366}
2367
2368static void handle_snapshot_monitor_command (const HChar *filename,
2369                                             Bool detailed)
2370{
2371   Snapshot snapshot;
2372
2373   if (!clo_pages_as_heap && !have_started_executing_code) {
2374      // See comments of variable have_started_executing_code.
2375      VG_(gdb_printf)
2376         ("error: cannot take snapshot before execution has started\n");
2377      return;
2378   }
2379
2380   clear_snapshot(&snapshot, /* do_sanity_check */ False);
2381   take_snapshot(&snapshot, Normal, get_time(), detailed);
2382   write_snapshots_to_file ((filename == NULL) ?
2383                            "massif.vgdb.out" : filename,
2384                            &snapshot,
2385                            1);
2386   delete_snapshot(&snapshot);
2387}
2388
2389static Bool handle_gdb_monitor_command (ThreadId tid, HChar *req)
2390{
2391   HChar* wcmd;
2392   HChar s[VG_(strlen(req)) + 1]; /* copy for strtok_r */
2393   HChar *ssaveptr;
2394
2395   VG_(strcpy) (s, req);
2396
2397   wcmd = VG_(strtok_r) (s, " ", &ssaveptr);
2398   switch (VG_(keyword_id) ("help snapshot detailed_snapshot",
2399                            wcmd, kwd_report_duplicated_matches)) {
2400   case -2: /* multiple matches */
2401      return True;
2402   case -1: /* not found */
2403      return False;
2404   case  0: /* help */
2405      print_monitor_help();
2406      return True;
2407   case  1: { /* snapshot */
2408      HChar* filename;
2409      filename = VG_(strtok_r) (NULL, " ", &ssaveptr);
2410      handle_snapshot_monitor_command (filename, False /* detailed */);
2411      return True;
2412   }
2413   case  2: { /* detailed_snapshot */
2414      HChar* filename;
2415      filename = VG_(strtok_r) (NULL, " ", &ssaveptr);
2416      handle_snapshot_monitor_command (filename, True /* detailed */);
2417      return True;
2418   }
2419   default:
2420      tl_assert(0);
2421      return False;
2422   }
2423}
2424
2425static void ms_print_stats (void)
2426{
2427#define STATS(format, args...) \
2428      VG_(dmsg)("Massif: " format, ##args)
2429
2430   STATS("heap allocs:           %u\n", n_heap_allocs);
2431   STATS("heap reallocs:         %u\n", n_heap_reallocs);
2432   STATS("heap frees:            %u\n", n_heap_frees);
2433   STATS("ignored heap allocs:   %u\n", n_ignored_heap_allocs);
2434   STATS("ignored heap frees:    %u\n", n_ignored_heap_frees);
2435   STATS("ignored heap reallocs: %u\n", n_ignored_heap_reallocs);
2436   STATS("stack allocs:          %u\n", n_stack_allocs);
2437   STATS("stack frees:           %u\n", n_stack_frees);
2438   STATS("XPts:                  %u\n", n_xpts);
2439   STATS("top-XPts:              %u (%d%%)\n",
2440      alloc_xpt->n_children,
2441      ( n_xpts ? alloc_xpt->n_children * 100 / n_xpts : 0));
2442   STATS("XPt init expansions:   %u\n", n_xpt_init_expansions);
2443   STATS("XPt later expansions:  %u\n", n_xpt_later_expansions);
2444   STATS("SXPt allocs:           %u\n", n_sxpt_allocs);
2445   STATS("SXPt frees:            %u\n", n_sxpt_frees);
2446   STATS("skipped snapshots:     %u\n", n_skipped_snapshots);
2447   STATS("real snapshots:        %u\n", n_real_snapshots);
2448   STATS("detailed snapshots:    %u\n", n_detailed_snapshots);
2449   STATS("peak snapshots:        %u\n", n_peak_snapshots);
2450   STATS("cullings:              %u\n", n_cullings);
2451   STATS("XCon redos:            %u\n", n_XCon_redos);
2452#undef STATS
2453}
2454
2455//------------------------------------------------------------//
2456//--- Finalisation                                         ---//
2457//------------------------------------------------------------//
2458
2459static void ms_fini(Int exit_status)
2460{
2461   // Output.
2462   write_snapshots_array_to_file();
2463
2464   // Stats
2465   tl_assert(n_xpts > 0);  // always have alloc_xpt
2466
2467   if (VG_(clo_stats))
2468      ms_print_stats();
2469}
2470
2471
2472//------------------------------------------------------------//
2473//--- Initialisation                                       ---//
2474//------------------------------------------------------------//
2475
2476static void ms_post_clo_init(void)
2477{
2478   Int i;
2479   HChar* LD_PRELOAD_val;
2480   HChar* s;
2481   HChar* s2;
2482
2483   // Check options.
2484   if (clo_pages_as_heap) {
2485      if (clo_stacks) {
2486         VG_(fmsg_bad_option)(
2487            "--pages-as-heap=yes together with --stacks=yes", "");
2488      }
2489   }
2490   if (!clo_heap) {
2491      clo_pages_as_heap = False;
2492   }
2493
2494   // If --pages-as-heap=yes we don't want malloc replacement to occur.  So we
2495   // disable vgpreload_massif-$PLATFORM.so by removing it from LD_PRELOAD (or
2496   // platform-equivalent).  We replace it entirely with spaces because then
2497   // the linker doesn't complain (it does complain if we just change the name
2498   // to a bogus file).  This is a bit of a hack, but LD_PRELOAD is setup well
2499   // before tool initialisation, so this seems the best way to do it.
2500   if (clo_pages_as_heap) {
2501      clo_heap_admin = 0;     // No heap admin on pages.
2502
2503      LD_PRELOAD_val = VG_(getenv)( VG_(LD_PRELOAD_var_name) );
2504      tl_assert(LD_PRELOAD_val);
2505
2506      // Make sure the vgpreload_core-$PLATFORM entry is there, for sanity.
2507      s2 = VG_(strstr)(LD_PRELOAD_val, "vgpreload_core");
2508      tl_assert(s2);
2509
2510      // Now find the vgpreload_massif-$PLATFORM entry.
2511      s2 = VG_(strstr)(LD_PRELOAD_val, "vgpreload_massif");
2512      tl_assert(s2);
2513
2514      // Blank out everything to the previous ':', which must be there because
2515      // of the preceding vgpreload_core-$PLATFORM entry.
2516      for (s = s2; *s != ':'; s--) {
2517         *s = ' ';
2518      }
2519
2520      // Blank out everything to the end of the entry, which will be '\0' if
2521      // LD_PRELOAD was empty before Valgrind started, or ':' otherwise.
2522      for (s = s2; *s != ':' && *s != '\0'; s++) {
2523         *s = ' ';
2524      }
2525   }
2526
2527   // Print alloc-fns and ignore-fns, if necessary.
2528   if (VG_(clo_verbosity) > 1) {
2529      VERB(1, "alloc-fns:\n");
2530      for (i = 0; i < VG_(sizeXA)(alloc_fns); i++) {
2531         HChar** fn_ptr = VG_(indexXA)(alloc_fns, i);
2532         VERB(1, "  %s\n", *fn_ptr);
2533      }
2534
2535      VERB(1, "ignore-fns:\n");
2536      if (0 == VG_(sizeXA)(ignore_fns)) {
2537         VERB(1, "  <empty>\n");
2538      }
2539      for (i = 0; i < VG_(sizeXA)(ignore_fns); i++) {
2540         HChar** fn_ptr = VG_(indexXA)(ignore_fns, i);
2541         VERB(1, "  %d: %s\n", i, *fn_ptr);
2542      }
2543   }
2544
2545   // Events to track.
2546   if (clo_stacks) {
2547      VG_(track_new_mem_stack)        ( new_mem_stack        );
2548      VG_(track_die_mem_stack)        ( die_mem_stack        );
2549      VG_(track_new_mem_stack_signal) ( new_mem_stack_signal );
2550      VG_(track_die_mem_stack_signal) ( die_mem_stack_signal );
2551   }
2552
2553   if (clo_pages_as_heap) {
2554      VG_(track_new_mem_startup) ( ms_new_mem_startup );
2555      VG_(track_new_mem_brk)     ( ms_new_mem_brk     );
2556      VG_(track_new_mem_mmap)    ( ms_new_mem_mmap    );
2557
2558      VG_(track_copy_mem_remap)  ( ms_copy_mem_remap  );
2559
2560      VG_(track_die_mem_brk)     ( ms_die_mem_brk     );
2561      VG_(track_die_mem_munmap)  ( ms_die_mem_munmap  );
2562   }
2563
2564   // Initialise snapshot array, and sanity-check it.
2565   snapshots = VG_(malloc)("ms.main.mpoci.1",
2566                           sizeof(Snapshot) * clo_max_snapshots);
2567   // We don't want to do snapshot sanity checks here, because they're
2568   // currently uninitialised.
2569   for (i = 0; i < clo_max_snapshots; i++) {
2570      clear_snapshot( & snapshots[i], /*do_sanity_check*/False );
2571   }
2572   sanity_check_snapshots_array();
2573}
2574
2575static void ms_pre_clo_init(void)
2576{
2577   VG_(details_name)            ("Massif");
2578   VG_(details_version)         (NULL);
2579   VG_(details_description)     ("a heap profiler");
2580   VG_(details_copyright_author)(
2581      "Copyright (C) 2003-2013, and GNU GPL'd, by Nicholas Nethercote");
2582   VG_(details_bug_reports_to)  (VG_BUGS_TO);
2583
2584   VG_(details_avg_translation_sizeB) ( 330 );
2585
2586   VG_(clo_vex_control).iropt_register_updates
2587      = VexRegUpdSpAtMemAccess; // overridable by the user.
2588
2589   // Basic functions.
2590   VG_(basic_tool_funcs)          (ms_post_clo_init,
2591                                   ms_instrument,
2592                                   ms_fini);
2593
2594   // Needs.
2595   VG_(needs_libc_freeres)();
2596   VG_(needs_command_line_options)(ms_process_cmd_line_option,
2597                                   ms_print_usage,
2598                                   ms_print_debug_usage);
2599   VG_(needs_client_requests)     (ms_handle_client_request);
2600   VG_(needs_sanity_checks)       (ms_cheap_sanity_check,
2601                                   ms_expensive_sanity_check);
2602   VG_(needs_print_stats)         (ms_print_stats);
2603   VG_(needs_malloc_replacement)  (ms_malloc,
2604                                   ms___builtin_new,
2605                                   ms___builtin_vec_new,
2606                                   ms_memalign,
2607                                   ms_calloc,
2608                                   ms_free,
2609                                   ms___builtin_delete,
2610                                   ms___builtin_vec_delete,
2611                                   ms_realloc,
2612                                   ms_malloc_usable_size,
2613                                   0 );
2614
2615   // HP_Chunks.
2616   malloc_list = VG_(HT_construct)( "Massif's malloc list" );
2617
2618   // Dummy node at top of the context structure.
2619   alloc_xpt = new_XPt(/*ip*/0, /*parent*/NULL);
2620
2621   // Initialise alloc_fns and ignore_fns.
2622   init_alloc_fns();
2623   init_ignore_fns();
2624
2625   // Initialise args_for_massif.
2626   args_for_massif = VG_(newXA)(VG_(malloc), "ms.main.mprci.1",
2627                                VG_(free), sizeof(HChar*));
2628}
2629
2630VG_DETERMINE_INTERFACE_VERSION(ms_pre_clo_init)
2631
2632//--------------------------------------------------------------------//
2633//--- end                                                          ---//
2634//--------------------------------------------------------------------//
2635