1Release 3.3.1 (4 June 2008)
2~~~~~~~~~~~~~~~~~~~~~~~~~~~
33.3.1 fixes a bunch of bugs in 3.3.0, adds support for glibc-2.8 based
4systems (openSUSE 11, Fedora Core 9), improves the existing glibc-2.7
5support, and adds support for the SSSE3 (Core 2) instruction set.
6
73.3.1 will likely be the last release that supports some very old
8systems.  In particular, the next major release, 3.4.0, will drop
9support for the old LinuxThreads threading library, and for gcc
10versions prior to 3.0.
11
12The fixed bugs are as follows.  Note that "n-i-bz" stands for "not in
13bugzilla" -- that is, a bug that was reported to us but never got a
14bugzilla entry.  We encourage you to file bugs in bugzilla
15(http://bugs.kde.org/enter_valgrind_bug.cgi) rather than mailing the
16developers (or mailing lists) directly -- bugs that are not entered
17into bugzilla tend to get forgotten about or ignored.
18
19n-i-bz  Massif segfaults at exit
20n-i-bz  Memcheck asserts on Altivec code
21n-i-bz  fix sizeof bug in Helgrind
22n-i-bz  check fd on sys_llseek
23n-i-bz  update syscall lists to kernel 2.6.23.1
24n-i-bz  support sys_sync_file_range
25n-i-bz  handle sys_sysinfo, sys_getresuid, sys_getresgid on ppc64-linux
26n-i-bz  intercept memcpy in 64-bit ld.so's
27n-i-bz  Fix wrappers for sys_{futimesat,utimensat}
28n-i-bz  Minor false-error avoidance fixes for Memcheck
29n-i-bz  libmpiwrap.c: add a wrapper for MPI_Waitany
30n-i-bz  helgrind support for glibc-2.8
31n-i-bz  partial fix for mc_leakcheck.c:698 assert:
32        'lc_shadows[i]->data + lc_shadows[i] ...
33n-i-bz  Massif/Cachegrind output corruption when programs fork
34n-i-bz  register allocator fix: handle spill stores correctly
35n-i-bz  add support for PA6T PowerPC CPUs
36126389  vex x86->IR: 0xF 0xAE (FXRSTOR)
37158525  ==126389
38152818  vex x86->IR: 0xF3 0xAC (repz lodsb)
39153196  vex x86->IR: 0xF2 0xA6 (repnz cmpsb)
40155011  vex x86->IR: 0xCF (iret)
41155091  Warning [...] unhandled DW_OP_ opcode 0x23
42156960  ==155901
43155528  support Core2/SSSE3 insns on x86/amd64
44155929  ms_print fails on massif outputs containing long lines
45157665  valgrind fails on shmdt(0) after shmat to 0
46157748  support x86 PUSHFW/POPFW
47158212  helgrind: handle pthread_rwlock_try{rd,wr}lock.
48158425  sys_poll incorrectly emulated when RES==0
49158744  vex amd64->IR: 0xF0 0x41 0xF 0xC0 (xaddb)
50160907  Support for a couple of recent Linux syscalls
51161285  Patch -- support for eventfd() syscall
52161378  illegal opcode in debug libm (FUCOMPP)
53160136  ==161378
54161487  number of suppressions files is limited to 10
55162386  ms_print typo in milliseconds time unit for massif
56161036  exp-drd: client allocated memory was never freed
57162663  signalfd_wrapper fails on 64bit linux
58
59(3.3.1.RC1:  2 June 2008, vex r1854, valgrind r8169).
60(3.3.1:      4 June 2008, vex r1854, valgrind r8180).
61
62
63
64Release 3.3.0 (7 December 2007)
65~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
663.3.0 is a feature release with many significant improvements and the
67usual collection of bug fixes.  This release supports X86/Linux,
68AMD64/Linux, PPC32/Linux and PPC64/Linux.  Support for recent distros
69(using gcc 4.3, glibc 2.6 and 2.7) has been added.
70
71The main excitement in 3.3.0 is new and improved tools.  Helgrind
72works again, Massif has been completely overhauled and much improved,
73Cachegrind now does branch-misprediction profiling, and a new category
74of experimental tools has been created, containing two new tools:
75Omega and DRD.  There are many other smaller improvements.  In detail:
76
77- Helgrind has been completely overhauled and works for the first time
78  since Valgrind 2.2.0.  Supported functionality is: detection of
79  misuses of the POSIX PThreads API, detection of potential deadlocks
80  resulting from cyclic lock dependencies, and detection of data
81  races.  Compared to the 2.2.0 Helgrind, the race detection algorithm
82  has some significant improvements aimed at reducing the false error
83  rate.  Handling of various kinds of corner cases has been improved.
84  Efforts have been made to make the error messages easier to
85  understand.  Extensive documentation is provided.
86
87- Massif has been completely overhauled.  Instead of measuring
88  space-time usage -- which wasn't always useful and many people found
89  confusing -- it now measures space usage at various points in the
90  execution, including the point of peak memory allocation.  Its
91  output format has also changed: instead of producing PostScript
92  graphs and HTML text, it produces a single text output (via the new
93  'ms_print' script) that contains both a graph and the old textual
94  information, but in a more compact and readable form.  Finally, the
95  new version should be more reliable than the old one, as it has been
96  tested more thoroughly.
97
98- Cachegrind has been extended to do branch-misprediction profiling.
99  Both conditional and indirect branches are profiled.  The default
100  behaviour of Cachegrind is unchanged.  To use the new functionality,
101  give the option --branch-sim=yes.
102
103- A new category of "experimental tools" has been created.  Such tools
104  may not work as well as the standard tools, but are included because
105  some people will find them useful, and because exposure to a wider
106  user group provides tool authors with more end-user feedback.  These
107  tools have a "exp-" prefix attached to their names to indicate their
108  experimental nature.  Currently there are two experimental tools:
109
110  * exp-Omega: an instantaneous leak detector.  See
111    exp-omega/docs/omega_introduction.txt.
112
113  * exp-DRD: a data race detector based on the happens-before
114    relation.  See exp-drd/docs/README.txt.
115
116- Scalability improvements for very large programs, particularly those
117  which have a million or more malloc'd blocks in use at once.  These
118  improvements mostly affect Memcheck.  Memcheck is also up to 10%
119  faster for all programs, with x86-linux seeing the largest
120  improvement.
121
122- Works well on the latest Linux distros.  Has been tested on Fedora
123  Core 8 (x86, amd64, ppc32, ppc64) and openSUSE 10.3.  glibc 2.6 and
124  2.7 are supported.  gcc-4.3 (in its current pre-release state) is
125  supported.  At the same time, 3.3.0 retains support for older
126  distros.
127
128- The documentation has been modestly reorganised with the aim of
129  making it easier to find information on common-usage scenarios.
130  Some advanced material has been moved into a new chapter in the main
131  manual, so as to unclutter the main flow, and other tidying up has
132  been done.
133
134- There is experimental support for AIX 5.3, both 32-bit and 64-bit
135  processes.  You need to be running a 64-bit kernel to use Valgrind
136  on a 64-bit executable.
137
138- There have been some changes to command line options, which may
139  affect you:
140
141  * --log-file-exactly and
142    --log-file-qualifier options have been removed.
143
144    To make up for this --log-file option has been made more powerful.
145    It now accepts a %p format specifier, which is replaced with the
146    process ID, and a %q{FOO} format specifier, which is replaced with
147    the contents of the environment variable FOO.
148
149  * --child-silent-after-fork=yes|no [no]
150
151    Causes Valgrind to not show any debugging or logging output for
152    the child process resulting from a fork() call.  This can make the
153    output less confusing (although more misleading) when dealing with
154    processes that create children.
155
156  * --cachegrind-out-file, --callgrind-out-file and --massif-out-file
157
158    These control the names of the output files produced by
159    Cachegrind, Callgrind and Massif.  They accept the same %p and %q
160    format specifiers that --log-file accepts.  --callgrind-out-file
161    replaces Callgrind's old --base option.
162
163  * Cachegrind's 'cg_annotate' script no longer uses the --<pid>
164    option to specify the output file.  Instead, the first non-option
165    argument is taken to be the name of the output file, and any
166    subsequent non-option arguments are taken to be the names of
167    source files to be annotated.
168
169  * Cachegrind and Callgrind now use directory names where possible in
170    their output files.  This means that the -I option to
171    'cg_annotate' and 'callgrind_annotate' should not be needed in
172    most cases.  It also means they can correctly handle the case
173    where two source files in different directories have the same
174    name.
175
176- Memcheck offers a new suppression kind: "Jump".  This is for
177  suppressing jump-to-invalid-address errors.  Previously you had to
178  use an "Addr1" suppression, which didn't make much sense.
179
180- Memcheck has new flags --malloc-fill=<hexnum> and
181  --free-fill=<hexnum> which free malloc'd / free'd areas with the
182  specified byte.  This can help shake out obscure memory corruption
183  problems.  The definedness and addressability of these areas is
184  unchanged -- only the contents are affected.
185
186- The behaviour of Memcheck's client requests VALGRIND_GET_VBITS and
187  VALGRIND_SET_VBITS have changed slightly.  They no longer issue
188  addressability errors -- if either array is partially unaddressable,
189  they just return 3 (as before).  Also, SET_VBITS doesn't report
190  definedness errors if any of the V bits are undefined.
191
192- The following Memcheck client requests have been removed:
193    VALGRIND_MAKE_NOACCESS
194    VALGRIND_MAKE_WRITABLE
195    VALGRIND_MAKE_READABLE
196    VALGRIND_CHECK_WRITABLE
197    VALGRIND_CHECK_READABLE
198    VALGRIND_CHECK_DEFINED
199  They were deprecated in 3.2.0, when equivalent but better-named client
200  requests were added.  See the 3.2.0 release notes for more details.
201
202- The behaviour of the tool Lackey has changed slightly.  First, the output
203  from --trace-mem has been made more compact, to reduce the size of the
204  traces.  Second, a new option --trace-superblocks has been added, which
205  shows the addresses of superblocks (code blocks) as they are executed.
206
207- The following bugs have been fixed.  Note that "n-i-bz" stands for
208  "not in bugzilla" -- that is, a bug that was reported to us but
209  never got a bugzilla entry.  We encourage you to file bugs in
210  bugzilla (http://bugs.kde.org/enter_valgrind_bug.cgi) rather than
211  mailing the developers (or mailing lists) directly.
212
213  n-i-bz  x86_linux_REDIR_FOR_index() broken
214  n-i-bz  guest-amd64/toIR.c:2512 (dis_op2_E_G): Assertion `0' failed.
215  n-i-bz  Support x86 INT insn (INT (0xCD) 0x40 - 0x43)
216  n-i-bz  Add sys_utimensat system call for Linux x86 platform
217   79844  Helgrind complains about race condition which does not exist
218   82871  Massif output function names too short
219   89061  Massif: ms_main.c:485 (get_XCon): Assertion `xpt->max_chi...'
220   92615  Write output from Massif at crash
221   95483  massif feature request: include peak allocation in report
222  112163  MASSIF crashed with signal 7 (SIGBUS) after running 2 days
223  119404  problems running setuid executables (partial fix)
224  121629  add instruction-counting mode for timing
225  127371  java vm giving unhandled instruction bytes: 0x26 0x2E 0x64 0x65
226  129937  ==150380
227  129576  Massif loses track of memory, incorrect graphs
228  132132  massif --format=html output does not do html entity escaping
229  132950  Heap alloc/usage summary
230  133962  unhandled instruction bytes: 0xF2 0x4C 0xF 0x10
231  134990  use -fno-stack-protector if possible
232  136382  ==134990
233  137396  I would really like helgrind to work again...
234  137714  x86/amd64->IR: 0x66 0xF 0xF7 0xC6 (maskmovq, maskmovdq)
235  141631  Massif: percentages don't add up correctly
236  142706  massif numbers don't seem to add up
237  143062  massif crashes on app exit with signal 8 SIGFPE
238  144453  (get_XCon): Assertion 'xpt->max_children != 0' failed.
239  145559  valgrind aborts when malloc_stats is called
240  145609  valgrind aborts all runs with 'repeated section!'
241  145622  --db-attach broken again on x86-64
242  145837  ==149519
243  145887  PPC32: getitimer() system call is not supported
244  146252  ==150678
245  146456  (update_XCon): Assertion 'xpt->curr_space >= -space_delta'...
246  146701  ==134990
247  146781  Adding support for private futexes
248  147325  valgrind internal error on syscall (SYS_io_destroy, 0)
249  147498  amd64->IR: 0xF0 0xF 0xB0 0xF (lock cmpxchg %cl,(%rdi))
250  147545  Memcheck: mc_main.c:817 (get_sec_vbits8): Assertion 'n' failed.
251  147628  SALC opcode 0xd6 unimplemented
252  147825  crash on amd64-linux with gcc 4.2 and glibc 2.6 (CFI)
253  148174  Incorrect type of freed_list_volume causes assertion [...]
254  148447  x86_64 : new NOP codes: 66 66 66 66 2e 0f 1f
255  149182  PPC Trap instructions not implemented in valgrind
256  149504  Assertion hit on alloc_xpt->curr_space >= -space_delta
257  149519  ppc32: V aborts with SIGSEGV on execution of a signal handler
258  149892  ==137714
259  150044  SEGV during stack deregister
260  150380  dwarf/gcc interoperation (dwarf3 read problems)
261  150408  ==148447
262  150678  guest-amd64/toIR.c:3741 (dis_Grp5): Assertion `sz == 4' failed
263  151209  V unable to execute programs for users with UID > 2^16
264  151938  help on --db-command= misleading
265  152022  subw $0x28, %%sp causes assertion failure in memcheck
266  152357  inb and outb not recognized in 64-bit mode
267  152501  vex x86->IR: 0x27 0x66 0x89 0x45 (daa)
268  152818  vex x86->IR: 0xF3 0xAC 0xFC 0x9C (rep lodsb)
269
270Developer-visible changes:
271
272- The names of some functions and types within the Vex IR have
273  changed.  Run 'svn log -r1689 VEX/pub/libvex_ir.h' for full details.
274  Any existing standalone tools will have to be updated to reflect
275  these changes.  The new names should be clearer.  The file
276  VEX/pub/libvex_ir.h is also much better commented.
277
278- A number of new debugging command line options have been added.
279  These are mostly of use for debugging the symbol table and line
280  number readers:
281
282  --trace-symtab-patt=<patt> limit debuginfo tracing to obj name <patt>
283  --trace-cfi=no|yes        show call-frame-info details? [no]
284  --debug-dump=syms         mimic /usr/bin/readelf --syms
285  --debug-dump=line         mimic /usr/bin/readelf --debug-dump=line
286  --debug-dump=frames       mimic /usr/bin/readelf --debug-dump=frames
287  --sym-offsets=yes|no      show syms in form 'name+offset' ? [no]
288
289- Internally, the code base has been further factorised and
290  abstractified, particularly with respect to support for non-Linux
291  OSs.
292
293(3.3.0.RC1:  2 Dec 2007, vex r1803, valgrind r7268).
294(3.3.0.RC2:  5 Dec 2007, vex r1804, valgrind r7282).
295(3.3.0.RC3:  9 Dec 2007, vex r1804, valgrind r7288).
296(3.3.0:     10 Dec 2007, vex r1804, valgrind r7290).
297
298
299
300Release 3.2.3 (29 Jan 2007)
301~~~~~~~~~~~~~~~~~~~~~~~~~~~
302Unfortunately 3.2.2 introduced a regression which can cause an
303assertion failure ("vex: the `impossible' happened: eqIRConst") when
304running obscure pieces of SSE code.  3.2.3 fixes this and adds one
305more glibc-2.5 intercept.  In all other respects it is identical to
3063.2.2.  Please do not use (or package) 3.2.2; instead use 3.2.3.
307
308n-i-bz   vex: the `impossible' happened: eqIRConst
309n-i-bz   Add an intercept for glibc-2.5 __stpcpy_chk
310
311(3.2.3: 29 Jan 2007, vex r1732, valgrind r6560).
312
313
314Release 3.2.2 (22 Jan 2007)
315~~~~~~~~~~~~~~~~~~~~~~~~~~~
3163.2.2 fixes a bunch of bugs in 3.2.1, adds support for glibc-2.5 based
317systems (openSUSE 10.2, Fedora Core 6), improves support for icc-9.X
318compiled code, and brings modest performance improvements in some
319areas, including amd64 floating point, powerpc support, and startup
320responsiveness on all targets.
321
322The fixed bugs are as follows.  Note that "n-i-bz" stands for "not in
323bugzilla" -- that is, a bug that was reported to us but never got a
324bugzilla entry.  We encourage you to file bugs in bugzilla
325(http://bugs.kde.org/enter_valgrind_bug.cgi) rather than mailing the
326developers (or mailing lists) directly.
327
328129390   ppc?->IR: some kind of VMX prefetch (dstt)
329129968   amd64->IR: 0xF 0xAE 0x0 (fxsave)
330134319   ==129968
331133054   'make install' fails with syntax errors
332118903   ==133054
333132998   startup fails in when running on UML
334134207   pkg-config output contains @VG_PLATFORM@
335134727   valgrind exits with "Value too large for defined data type"
336n-i-bz   ppc32/64: support mcrfs
337n-i-bz   Cachegrind/Callgrind: Update cache parameter detection
338135012   x86->IR: 0xD7 0x8A 0xE0 0xD0 (xlat)
339125959   ==135012
340126147   x86->IR: 0xF2 0xA5 0xF 0x77 (repne movsw)
341136650   amd64->IR: 0xC2 0x8 0x0
342135421   x86->IR: unhandled Grp5(R) case 6
343n-i-bz   Improved documentation of the IR intermediate representation
344n-i-bz   jcxz (x86) (users list, 8 Nov)
345n-i-bz   ExeContext hashing fix
346n-i-bz   fix CFI reading failures ("Dwarf CFI 0:24 0:32 0:48 0:7")
347n-i-bz   fix Cachegrind/Callgrind simulation bug
348n-i-bz   libmpiwrap.c: fix handling of MPI_LONG_DOUBLE
349n-i-bz   make User errors suppressible
350136844   corrupted malloc line when using --gen-suppressions=yes
351138507   ==136844
352n-i-bz   Speed up the JIT's register allocator
353n-i-bz   Fix confusing leak-checker flag hints
354n-i-bz   Support recent autoswamp versions
355n-i-bz   ppc32/64 dispatcher speedups
356n-i-bz   ppc64 front end rld/rlw improvements
357n-i-bz   ppc64 back end imm64 improvements
358136300   support 64K pages on ppc64-linux
359139124   == 136300
360n-i-bz   fix ppc insn set tests for gcc >= 4.1
361137493   x86->IR: recent binutils no-ops
362137714   x86->IR: 0x66 0xF 0xF7 0xC6 (maskmovdqu)
363138424   "failed in UME with error 22" (produce a better error msg)
364138856   ==138424
365138627   Enhancement support for prctl ioctls
366138896   Add support for usb ioctls
367136059   ==138896
368139050   ppc32->IR: mfspr 268/269 instructions not handled
369n-i-bz   ppc32->IR: lvxl/stvxl
370n-i-bz   glibc-2.5 support
371n-i-bz   memcheck: provide replacement for mempcpy
372n-i-bz   memcheck: replace bcmp in ld.so
373n-i-bz   Use 'ifndef' in VEX's Makefile correctly
374n-i-bz   Suppressions for MVL 4.0.1 on ppc32-linux
375n-i-bz   libmpiwrap.c: Fixes for MPICH
376n-i-bz   More robust handling of hinted client mmaps
377139776   Invalid read in unaligned memcpy with Intel compiler v9
378n-i-bz   Generate valid XML even for very long fn names
379n-i-bz   Don't prompt about suppressions for unshown reachable leaks
380139910   amd64 rcl is not supported
381n-i-bz   DWARF CFI reader: handle DW_CFA_undefined
382n-i-bz   DWARF CFI reader: handle icc9 generated CFI info better
383n-i-bz   fix false uninit-value errs in icc9 generated FP code
384n-i-bz   reduce extraneous frames in libmpiwrap.c
385n-i-bz   support pselect6 on amd64-linux
386
387(3.2.2: 22 Jan 2007, vex r1729, valgrind r6545).
388
389
390Release 3.2.1 (16 Sept 2006)
391~~~~~~~~~~~~~~~~~~~~~~~~~~~~
3923.2.1 adds x86/amd64 support for all SSE3 instructions except monitor
393and mwait, further reduces memcheck's false error rate on all
394platforms, adds support for recent binutils (in OpenSUSE 10.2 and
395Fedora Rawhide) and fixes a bunch of bugs in 3.2.0.  Some of the fixed
396bugs were causing large programs to segfault with --tool=callgrind and
397--tool=cachegrind, so an upgrade is recommended.
398
399In view of the fact that any 3.3.0 release is unlikely to happen until
400well into 1Q07, we intend to keep the 3.2.X line alive for a while
401yet, and so we tentatively plan a 3.2.2 release sometime in December
40206.
403
404The fixed bugs are as follows.  Note that "n-i-bz" stands for "not in
405bugzilla" -- that is, a bug that was reported to us but never got a
406bugzilla entry.
407
408n-i-bz   Expanding brk() into last available page asserts
409n-i-bz   ppc64-linux stack RZ fast-case snafu
410n-i-bz   'c' in --gen-supps=yes doesn't work
411n-i-bz   VG_N_SEGMENTS too low (users, 28 June)
412n-i-bz   VG_N_SEGNAMES too low (Stu Robinson)
413106852   x86->IR: fisttp (SSE3)
414117172   FUTEX_WAKE does not use uaddr2
415124039   Lacks support for VKI_[GP]IO_UNIMAP*
416127521   amd64->IR: 0xF0 0x48 0xF 0xC7 (cmpxchg8b)
417128917   amd64->IR: 0x66 0xF 0xF6 0xC4 (psadbw,SSE2)
418129246   JJ: ppc32/ppc64 syscalls, w/ patch
419129358   x86->IR: fisttpl (SSE3)
420129866   cachegrind/callgrind causes executable to die
421130020   Can't stat .so/.exe error while reading symbols
422130388   Valgrind aborts when process calls malloc_trim()
423130638   PATCH: ppc32 missing system calls
424130785   amd64->IR: unhandled instruction "pushfq"
425131481:  (HINT_NOP) vex x86->IR: 0xF 0x1F 0x0 0xF
426131298   ==131481
427132146   Programs with long sequences of bswap[l,q]s
428132918   vex amd64->IR: 0xD9 0xF8 (fprem)
429132813   Assertion at priv/guest-x86/toIR.c:652 fails
430133051   'cfsi->len > 0 && cfsi->len < 2000000' failed
431132722   valgrind header files are not standard C
432n-i-bz   Livelocks entire machine (users list, Timothy Terriberry)
433n-i-bz   Alex Bennee mmap problem (9 Aug)
434n-i-bz   BartV: Don't print more lines of a stack-trace than were obtained.
435n-i-bz   ppc32 SuSE 10.1 redir
436n-i-bz   amd64 padding suppressions
437n-i-bz   amd64 insn printing fix.
438n-i-bz   ppc cmp reg,reg fix
439n-i-bz   x86/amd64 iropt e/rflag reduction rules
440n-i-bz   SuSE 10.1 (ppc32) minor fixes
441133678   amd64->IR: 0x48 0xF 0xC5 0xC0 (pextrw?)
442133694   aspacem assertion: aspacem_minAddr <= holeStart
443n-i-bz   callgrind: fix warning about malformed creator line
444n-i-bz   callgrind: fix annotate script for data produced with
445         --dump-instr=yes
446n-i-bz   callgrind: fix failed assertion when toggling
447         instrumentation mode
448n-i-bz   callgrind: fix annotate script fix warnings with
449         --collect-jumps=yes
450n-i-bz   docs path hardwired (Dennis Lubert)
451
452The following bugs were not fixed, due primarily to lack of developer
453time, and also because bug reporters did not answer requests for
454feedback in time for the release:
455
456129390   ppc?->IR: some kind of VMX prefetch (dstt)
457129968   amd64->IR: 0xF 0xAE 0x0 (fxsave)
458133054   'make install' fails with syntax errors
459n-i-bz   Signal race condition (users list, 13 June, Johannes Berg)
460n-i-bz   Unrecognised instruction at address 0x70198EC2 (users list,
461         19 July, Bennee)
462132998   startup fails in when running on UML
463
464The following bug was tentatively fixed on the mainline but the fix
465was considered too risky to push into 3.2.X:
466
467133154   crash when using client requests to register/deregister stack
468
469(3.2.1: 16 Sept 2006, vex r1658, valgrind r6070).
470
471
472Release 3.2.0 (7 June 2006)
473~~~~~~~~~~~~~~~~~~~~~~~~~~~
4743.2.0 is a feature release with many significant improvements and the
475usual collection of bug fixes.  This release supports X86/Linux,
476AMD64/Linux, PPC32/Linux and PPC64/Linux.
477
478Performance, especially of Memcheck, is improved, Addrcheck has been
479removed, Callgrind has been added, PPC64/Linux support has been added,
480Lackey has been improved, and MPI support has been added.  In detail:
481
482- Memcheck has improved speed and reduced memory use.  Run times are
483  typically reduced by 15-30%, averaging about 24% for SPEC CPU2000.
484  The other tools have smaller but noticeable speed improvements.  We
485  are interested to hear what improvements users get.
486
487  Memcheck uses less memory due to the introduction of a compressed
488  representation for shadow memory.  The space overhead has been
489  reduced by a factor of up to four, depending on program behaviour.
490  This means you should be able to run programs that use more memory
491  than before without hitting problems.
492
493- Addrcheck has been removed.  It has not worked since version 2.4.0,
494  and the speed and memory improvements to Memcheck make it redundant.
495  If you liked using Addrcheck because it didn't give undefined value
496  errors, you can use the new Memcheck option --undef-value-errors=no
497  to get the same behaviour.
498
499- The number of undefined-value errors incorrectly reported by
500  Memcheck has been reduced (such false reports were already very
501  rare).  In particular, efforts have been made to ensure Memcheck
502  works really well with gcc 4.0/4.1-generated code on X86/Linux and
503  AMD64/Linux.
504
505- Josef Weidendorfer's popular Callgrind tool has been added.  Folding
506  it in was a logical step given its popularity and usefulness, and
507  makes it easier for us to ensure it works "out of the box" on all
508  supported targets.  The associated KDE KCachegrind GUI remains a
509  separate project.
510
511- A new release of the Valkyrie GUI for Memcheck, version 1.2.0,
512  accompanies this release.  Improvements over previous releases
513  include improved robustness, many refinements to the user interface,
514  and use of a standard autoconf/automake build system.  You can get
515  it from http://www.valgrind.org/downloads/guis.html.
516
517- Valgrind now works on PPC64/Linux.  As with the AMD64/Linux port,
518  this supports programs using to 32G of address space.  On 64-bit
519  capable PPC64/Linux setups, you get a dual architecture build so
520  that both 32-bit and 64-bit executables can be run.  Linux on POWER5
521  is supported, and POWER4 is also believed to work.  Both 32-bit and
522  64-bit DWARF2 is supported.  This port is known to work well with
523  both gcc-compiled and xlc/xlf-compiled code.
524
525- Floating point accuracy has been improved for PPC32/Linux.
526  Specifically, the floating point rounding mode is observed on all FP
527  arithmetic operations, and multiply-accumulate instructions are
528  preserved by the compilation pipeline.  This means you should get FP
529  results which are bit-for-bit identical to a native run.  These
530  improvements are also present in the PPC64/Linux port.
531
532- Lackey, the example tool, has been improved:
533
534  * It has a new option --detailed-counts (off by default) which
535    causes it to print out a count of loads, stores and ALU operations
536    done, and their sizes.
537
538  * It has a new option --trace-mem (off by default) which causes it
539    to print out a trace of all memory accesses performed by a
540    program.  It's a good starting point for building Valgrind tools
541    that need to track memory accesses.  Read the comments at the top
542    of the file lackey/lk_main.c for details.
543
544  * The original instrumentation (counting numbers of instructions,
545    jumps, etc) is now controlled by a new option --basic-counts.  It
546    is on by default.
547
548- MPI support: partial support for debugging distributed applications
549  using the MPI library specification has been added.  Valgrind is
550  aware of the memory state changes caused by a subset of the MPI
551  functions, and will carefully check data passed to the (P)MPI_
552  interface.
553
554- A new flag, --error-exitcode=, has been added.  This allows changing
555  the exit code in runs where Valgrind reported errors, which is
556  useful when using Valgrind as part of an automated test suite.
557
558- Various segfaults when reading old-style "stabs" debug information
559  have been fixed.
560
561- A simple performance evaluation suite has been added.  See
562  perf/README and README_DEVELOPERS for details.  There are
563  various bells and whistles.
564
565- New configuration flags:
566    --enable-only32bit
567    --enable-only64bit
568  By default, on 64 bit platforms (ppc64-linux, amd64-linux) the build
569  system will attempt to build a Valgrind which supports both 32-bit
570  and 64-bit executables.  This may not be what you want, and you can
571  override the default behaviour using these flags.
572
573Please note that Helgrind is still not working.  We have made an
574important step towards making it work again, however, with the
575addition of function wrapping (see below).
576
577Other user-visible changes:
578
579- Valgrind now has the ability to intercept and wrap arbitrary
580  functions.  This is a preliminary step towards making Helgrind work
581  again, and was required for MPI support.
582
583- There are some changes to Memcheck's client requests.  Some of them
584  have changed names:
585
586    MAKE_NOACCESS  --> MAKE_MEM_NOACCESS
587    MAKE_WRITABLE  --> MAKE_MEM_UNDEFINED
588    MAKE_READABLE  --> MAKE_MEM_DEFINED
589
590    CHECK_WRITABLE --> CHECK_MEM_IS_ADDRESSABLE
591    CHECK_READABLE --> CHECK_MEM_IS_DEFINED
592    CHECK_DEFINED  --> CHECK_VALUE_IS_DEFINED
593
594  The reason for the change is that the old names are subtly
595  misleading.  The old names will still work, but they are deprecated
596  and may be removed in a future release.
597
598  We also added a new client request:
599
600    MAKE_MEM_DEFINED_IF_ADDRESSABLE(a, len)
601
602  which is like MAKE_MEM_DEFINED but only affects a byte if the byte is
603  already addressable.
604
605- The way client requests are encoded in the instruction stream has
606  changed.  Unfortunately, this means 3.2.0 will not honour client
607  requests compiled into binaries using headers from earlier versions
608  of Valgrind.  We will try to keep the client request encodings more
609  stable in future.
610
611BUGS FIXED:
612
613108258   NPTL pthread cleanup handlers not called
614117290   valgrind is sigKILL'd on startup
615117295   == 117290
616118703   m_signals.c:1427 Assertion 'tst->status == VgTs_WaitSys'
617118466   add %reg, %reg generates incorrect validity for bit 0
618123210   New: strlen from ld-linux on amd64
619123244   DWARF2 CFI reader: unhandled CFI instruction 0:18
620123248   syscalls in glibc-2.4: openat, fstatat, symlinkat
621123258   socketcall.recvmsg(msg.msg_iov[i] points to uninit
622123535   mremap(new_addr) requires MREMAP_FIXED in 4th arg
623123836   small typo in the doc
624124029   ppc compile failed: `vor' gcc 3.3.5
625124222   Segfault: @@don't know what type ':' is
626124475   ppc32: crash (syscall?) timer_settime()
627124499   amd64->IR: 0xF 0xE 0x48 0x85 (femms)
628124528   FATAL: aspacem assertion failed: segment_is_sane
629124697   vex x86->IR: 0xF 0x70 0xC9 0x0 (pshufw)
630124892   vex x86->IR: 0xF3 0xAE (REPx SCASB)
631126216   == 124892
632124808   ppc32: sys_sched_getaffinity() not handled
633n-i-bz   Very long stabs strings crash m_debuginfo
634n-i-bz   amd64->IR: 0x66 0xF 0xF5 (pmaddwd)
635125492   ppc32: support a bunch more syscalls
636121617   ppc32/64: coredumping gives assertion failure
637121814   Coregrind return error as exitcode patch
638126517   == 121814
639125607   amd64->IR: 0x66 0xF 0xA3 0x2 (btw etc)
640125651   amd64->IR: 0xF8 0x49 0xFF 0xE3 (clc?)
641126253   x86 movx is wrong
642126451   3.2 SVN doesn't work on ppc32 CPU's without FPU
643126217   increase # threads
644126243   vex x86->IR: popw mem
645126583   amd64->IR: 0x48 0xF 0xA4 0xC2 (shld $1,%rax,%rdx)
646126668   amd64->IR: 0x1C 0xFF (sbb $0xff,%al)
647126696   support for CDROMREADRAW ioctl and CDROMREADTOCENTRY fix
648126722   assertion: segment_is_sane at m_aspacemgr/aspacemgr.c:1624
649126938   bad checking for syscalls linkat, renameat, symlinkat
650
651(3.2.0RC1: 27 May  2006, vex r1626, valgrind r5947).
652(3.2.0:     7 June 2006, vex r1628, valgrind r5957).
653
654
655Release 3.1.1 (15 March 2006)
656~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
6573.1.1 fixes a bunch of bugs reported in 3.1.0.  There is no new
658functionality.  The fixed bugs are:
659
660(note: "n-i-bz" means "not in bugzilla" -- this bug does not have
661 a bugzilla entry).
662
663n-i-bz   ppc32: fsub 3,3,3 in dispatcher doesn't clear NaNs
664n-i-bz   ppc32: __NR_{set,get}priority
665117332   x86: missing line info with icc 8.1
666117366   amd64: 0xDD 0x7C fnstsw
667118274   == 117366
668117367   amd64: 0xD9 0xF4 fxtract
669117369   amd64: __NR_getpriority (140)
670117419   ppc32: lfsu f5, -4(r11)
671117419   ppc32: fsqrt
672117936   more stabs problems (segfaults while reading debug info)
673119914   == 117936
674120345   == 117936
675118239   amd64: 0xF 0xAE 0x3F (clflush)
676118939   vm86old system call
677n-i-bz   memcheck/tests/mempool reads freed memory
678n-i-bz   AshleyP's custom-allocator assertion
679n-i-bz   Dirk strict-aliasing stuff
680n-i-bz   More space for debugger cmd line (Dan Thaler)
681n-i-bz   Clarified leak checker output message
682n-i-bz   AshleyP's --gen-suppressions output fix
683n-i-bz   cg_annotate's --sort option broken
684n-i-bz   OSet 64-bit fastcmp bug
685n-i-bz   VG_(getgroups) fix (Shinichi Noda)
686n-i-bz   ppc32: allocate from callee-saved FP/VMX regs
687n-i-bz   misaligned path word-size bug in mc_main.c
688119297   Incorrect error message for sse code
689120410   x86: prefetchw (0xF 0xD 0x48 0x4)
690120728   TIOCSERGETLSR, TIOCGICOUNT, HDIO_GET_DMA ioctls
691120658   Build fixes for gcc 2.96
692120734   x86: Support for changing EIP in signal handler
693n-i-bz   memcheck/tests/zeropage de-looping fix
694n-i-bz   x86: fxtract doesn't work reliably
695121662   x86: lock xadd (0xF0 0xF 0xC0 0x2)
696121893   calloc does not always return zeroed memory
697121901   no support for syscall tkill
698n-i-bz   Suppression update for Debian unstable
699122067   amd64: fcmovnu (0xDB 0xD9)
700n-i-bz   ppc32: broken signal handling in cpu feature detection
701n-i-bz   ppc32: rounding mode problems (improved, partial fix only)
702119482   ppc32: mtfsb1
703n-i-bz   ppc32: mtocrf/mfocrf
704
705(3.1.1:  15 March 2006, vex r1597, valgrind r5771).
706
707
708Release 3.1.0 (25 November 2005)
709~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
7103.1.0 is a feature release with a number of significant improvements:
711AMD64 support is much improved, PPC32 support is good enough to be
712usable, and the handling of memory management and address space is
713much more robust.  In detail:
714
715- AMD64 support is much improved.  The 64-bit vs. 32-bit issues in
716  3.0.X have been resolved, and it should "just work" now in all
717  cases.  On AMD64 machines both 64-bit and 32-bit versions of
718  Valgrind are built.  The right version will be invoked
719  automatically, even when using --trace-children and mixing execution
720  between 64-bit and 32-bit executables.  Also, many more instructions
721  are supported.
722
723- PPC32 support is now good enough to be usable.  It should work with
724  all tools, but please let us know if you have problems.  Three
725  classes of CPUs are supported: integer only (no FP, no Altivec),
726  which covers embedded PPC uses, integer and FP but no Altivec
727  (G3-ish), and CPUs capable of Altivec too (G4, G5).
728
729- Valgrind's address space management has been overhauled.  As a
730  result, Valgrind should be much more robust with programs that use
731  large amounts of memory.  There should be many fewer "memory
732  exhausted" messages, and debug symbols should be read correctly on
733  large (eg. 300MB+) executables.  On 32-bit machines the full address
734  space available to user programs (usually 3GB or 4GB) can be fully
735  utilised.  On 64-bit machines up to 32GB of space is usable; when
736  using Memcheck that means your program can use up to about 14GB.
737
738  A side effect of this change is that Valgrind is no longer protected
739  against wild writes by the client.  This feature was nice but relied
740  on the x86 segment registers and so wasn't portable.
741
742- Most users should not notice, but as part of the address space
743  manager change, the way Valgrind is built has been changed.  Each
744  tool is now built as a statically linked stand-alone executable,
745  rather than as a shared object that is dynamically linked with the
746  core.  The "valgrind" program invokes the appropriate tool depending
747  on the --tool option.  This slightly increases the amount of disk
748  space used by Valgrind, but it greatly simplified many things and
749  removed Valgrind's dependence on glibc.
750
751Please note that Addrcheck and Helgrind are still not working.  Work
752is underway to reinstate them (or equivalents).  We apologise for the
753inconvenience.
754
755Other user-visible changes:
756
757- The --weird-hacks option has been renamed --sim-hints.
758
759- The --time-stamp option no longer gives an absolute date and time.
760  It now prints the time elapsed since the program began.
761
762- It should build with gcc-2.96.
763
764- Valgrind can now run itself (see README_DEVELOPERS for how).
765  This is not much use to you, but it means the developers can now
766  profile Valgrind using Cachegrind.  As a result a couple of
767  performance bad cases have been fixed.
768
769- The XML output format has changed slightly.  See
770  docs/internals/xml-output.txt.
771
772- Core dumping has been reinstated (it was disabled in 3.0.0 and 3.0.1).
773  If your program crashes while running under Valgrind, a core file with
774  the name "vgcore.<pid>" will be created (if your settings allow core
775  file creation).  Note that the floating point information is not all
776  there.  If Valgrind itself crashes, the OS will create a normal core
777  file.
778
779The following are some user-visible changes that occurred in earlier
780versions that may not have been announced, or were announced but not
781widely noticed.  So we're mentioning them now.
782
783- The --tool flag is optional once again;  if you omit it, Memcheck
784  is run by default.
785
786- The --num-callers flag now has a default value of 12.  It was
787  previously 4.
788
789- The --xml=yes flag causes Valgrind's output to be produced in XML
790  format.  This is designed to make it easy for other programs to
791  consume Valgrind's output.  The format is described in the file
792  docs/internals/xml-format.txt.
793
794- The --gen-suppressions flag supports an "all" value that causes every
795  suppression to be printed without asking.
796
797- The --log-file option no longer puts "pid" in the filename, eg. the
798  old name "foo.pid12345" is now "foo.12345".
799
800- There are several graphical front-ends for Valgrind, such as Valkyrie,
801  Alleyoop and Valgui.  See http://www.valgrind.org/downloads/guis.html
802  for a list.
803
804BUGS FIXED:
805
806109861  amd64 hangs at startup
807110301  ditto
808111554  valgrind crashes with Cannot allocate memory
809111809  Memcheck tool doesn't start java
810111901  cross-platform run of cachegrind fails on opteron
811113468  (vgPlain_mprotect_range): Assertion 'r != -1' failed.
812 92071  Reading debugging info uses too much memory
813109744  memcheck loses track of mmap from direct ld-linux.so.2
814110183  tail of page with _end
815 82301  FV memory layout too rigid
816 98278  Infinite recursion possible when allocating memory
817108994  Valgrind runs out of memory due to 133x overhead
818115643  valgrind cannot allocate memory
819105974  vg_hashtable.c static hash table
820109323  ppc32: dispatch.S uses Altivec insn, which doesn't work on POWER.
821109345  ptrace_setregs not yet implemented for ppc
822110831  Would like to be able to run against both 32 and 64 bit
823        binaries on AMD64
824110829  == 110831
825111781  compile of valgrind-3.0.0 fails on my linux (gcc 2.X prob)
826112670  Cachegrind: cg_main.c:486 (handleOneStatement ...
827112941  vex x86: 0xD9 0xF4 (fxtract)
828110201  == 112941
829113015  vex amd64->IR: 0xE3 0x14 0x48 0x83 (jrcxz)
830113126  Crash with binaries built with -gstabs+/-ggdb
831104065  == 113126
832115741  == 113126
833113403  Partial SSE3 support on x86
834113541  vex: Grp5(x86) (alt encoding inc/dec) case 1
835113642  valgrind crashes when trying to read debug information
836113810  vex x86->IR: 66 0F F6 (66 + PSADBW == SSE PSADBW)
837113796  read() and write() do not work if buffer is in shared memory
838113851  vex x86->IR: (pmaddwd): 0x66 0xF 0xF5 0xC7
839114366  vex amd64 cannnot handle __asm__( "fninit" )
840114412  vex amd64->IR: 0xF 0xAD 0xC2 0xD3 (128-bit shift, shrdq?)
841114455  vex amd64->IR: 0xF 0xAC 0xD0 0x1 (also shrdq)
842115590: amd64->IR: 0x67 0xE3 0x9 0xEB (address size override)
843115953  valgrind svn r5042 does not build with parallel make (-j3)
844116057  maximum instruction size - VG_MAX_INSTR_SZB too small?
845116483  shmat failes with invalid argument
846102202  valgrind crashes when realloc'ing until out of memory
847109487  == 102202
848110536  == 102202
849112687  == 102202
850111724  vex amd64->IR: 0x41 0xF 0xAB (more BT{,S,R,C} fun n games)
851111748  vex amd64->IR: 0xDD 0xE2 (fucom)
852111785  make fails if CC contains spaces
853111829  vex x86->IR: sbb AL, Ib
854111851  vex x86->IR: 0x9F 0x89 (lahf/sahf)
855112031  iopl on AMD64 and README_MISSING_SYSCALL_OR_IOCTL update
856112152  code generation for Xin_MFence on x86 with SSE0 subarch
857112167  == 112152
858112789  == 112152
859112199  naked ar tool is used in vex makefile
860112501  vex x86->IR: movq (0xF 0x7F 0xC1 0xF) (mmx MOVQ)
861113583  == 112501
862112538  memalign crash
863113190  Broken links in docs/html/
864113230  Valgrind sys_pipe on x86-64 wrongly thinks file descriptors
865        should be 64bit
866113996  vex amd64->IR: fucomp (0xDD 0xE9)
867114196  vex x86->IR: out %eax,(%dx) (0xEF 0xC9 0xC3 0x90)
868114289  Memcheck fails to intercept malloc when used in an uclibc environment
869114756  mbind syscall support
870114757  Valgrind dies with assertion: Assertion 'noLargerThan > 0' failed
871114563  stack tracking module not informed when valgrind switches threads
872114564  clone() and stacks
873114565  == 114564
874115496  glibc crashes trying to use sysinfo page
875116200  enable fsetxattr, fgetxattr, and fremovexattr for amd64
876
877(3.1.0RC1: 20 November 2005, vex r1466, valgrind r5224).
878(3.1.0:    26 November 2005, vex r1471, valgrind r5235).
879
880
881Release 3.0.1 (29 August 2005)
882~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
8833.0.1 fixes a bunch of bugs reported in 3.0.0.  There is no new
884functionality.  Some of the fixed bugs are critical, so if you
885use/distribute 3.0.0, an upgrade to 3.0.1 is recommended.  The fixed
886bugs are:
887
888(note: "n-i-bz" means "not in bugzilla" -- this bug does not have
889 a bugzilla entry).
890
891109313  (== 110505) x86 cmpxchg8b
892n-i-bz  x86: track but ignore changes to %eflags.AC (alignment check)
893110102  dis_op2_E_G(amd64)
894110202  x86 sys_waitpid(#286)
895110203  clock_getres(,0)
896110208  execve fail wrong retval
897110274  SSE1 now mandatory for x86
898110388  amd64 0xDD 0xD1
899110464  amd64 0xDC 0x1D FCOMP
900110478  amd64 0xF 0xD PREFETCH
901n-i-bz  XML <unique> printing wrong
902n-i-bz  Dirk r4359 (amd64 syscalls from trunk)
903110591  amd64 and x86: rdtsc not implemented properly
904n-i-bz  Nick r4384 (stub implementations of Addrcheck and Helgrind)
905110652  AMD64 valgrind crashes on cwtd instruction
906110653  AMD64 valgrind crashes on sarb $0x4,foo(%rip) instruction
907110656  PATH=/usr/bin::/bin valgrind foobar stats ./fooba
908110657  Small test fixes
909110671  vex x86->IR: unhandled instruction bytes: 0xF3 0xC3 (rep ret)
910n-i-bz  Nick (Cachegrind should not assert when it encounters a client
911        request.)
912110685  amd64->IR: unhandled instruction bytes: 0xE1 0x56 (loope Jb)
913110830  configuring with --host fails to build 32 bit on 64 bit target
914110875  Assertion when execve fails
915n-i-bz  Updates to Memcheck manual
916n-i-bz  Fixed broken malloc_usable_size()
917110898  opteron instructions missing: btq btsq btrq bsfq
918110954  x86->IR: unhandled instruction bytes: 0xE2 0xF6 (loop Jb)
919n-i-bz  Make suppressions work for "???" lines in stacktraces.
920111006  bogus warnings from linuxthreads
921111092  x86: dis_Grp2(Reg): unhandled case(x86)
922111231  sctp_getladdrs() and sctp_getpaddrs() returns uninitialized
923        memory
924111102  (comment #4)   Fixed 64-bit unclean "silly arg" message
925n-i-bz  vex x86->IR: unhandled instruction bytes: 0x14 0x0
926n-i-bz  minor umount/fcntl wrapper fixes
927111090  Internal Error running Massif
928101204  noisy warning
929111513  Illegal opcode for SSE instruction (x86 movups)
930111555  VEX/Makefile: CC is set to gcc
931n-i-bz  Fix XML bugs in FAQ
932
933(3.0.1: 29 August 05,
934        vex/branches/VEX_3_0_BRANCH r1367,
935        valgrind/branches/VALGRIND_3_0_BRANCH r4574).
936
937
938
939Release 3.0.0 (3 August 2005)
940~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
9413.0.0 is a major overhaul of Valgrind.  The most significant user
942visible change is that Valgrind now supports architectures other than
943x86.  The new architectures it supports are AMD64 and PPC32, and the
944infrastructure is present for other architectures to be added later.
945
946AMD64 support works well, but has some shortcomings:
947
948- It generally won't be as solid as the x86 version.  For example,
949  support for more obscure instructions and system calls may be missing.
950  We will fix these as they arise.
951
952- Address space may be limited; see the point about
953  position-independent executables below.
954
955- If Valgrind is built on an AMD64 machine, it will only run 64-bit
956  executables.  If you want to run 32-bit x86 executables under Valgrind
957  on an AMD64, you will need to build Valgrind on an x86 machine and
958  copy it to the AMD64 machine.  And it probably won't work if you do
959  something tricky like exec'ing a 32-bit program from a 64-bit program
960  while using --trace-children=yes.  We hope to improve this situation
961  in the future.
962
963The PPC32 support is very basic.  It may not work reliably even for
964small programs, but it's a start.  Many thanks to Paul Mackerras for
965his great work that enabled this support.  We are working to make
966PPC32 usable as soon as possible.
967
968Other user-visible changes:
969
970- Valgrind is no longer built by default as a position-independent
971  executable (PIE), as this caused too many problems.
972
973  Without PIE enabled, AMD64 programs will only be able to access 2GB of
974  address space.  We will fix this eventually, but not for the moment.
975
976  Use --enable-pie at configure-time to turn this on.
977
978- Support for programs that use stack-switching has been improved.  Use
979  the --max-stackframe flag for simple cases, and the
980  VALGRIND_STACK_REGISTER, VALGRIND_STACK_DEREGISTER and
981  VALGRIND_STACK_CHANGE client requests for trickier cases.
982
983- Support for programs that use self-modifying code has been improved,
984  in particular programs that put temporary code fragments on the stack.
985  This helps for C programs compiled with GCC that use nested functions,
986  and also Ada programs.  This is controlled with the --smc-check
987  flag, although the default setting should work in most cases.
988
989- Output can now be printed in XML format.  This should make it easier
990  for tools such as GUI front-ends and automated error-processing
991  schemes to use Valgrind output as input.  The --xml flag controls this.
992  As part of this change, ELF directory information is read from executables,
993  so absolute source file paths are available if needed.
994
995- Programs that allocate many heap blocks may run faster, due to
996  improvements in certain data structures.
997
998- Addrcheck is currently not working.  We hope to get it working again
999  soon.  Helgrind is still not working, as was the case for the 2.4.0
1000  release.
1001
1002- The JITter has been completely rewritten, and is now in a separate
1003  library, called Vex.  This enabled a lot of the user-visible changes,
1004  such as new architecture support.  The new JIT unfortunately translates
1005  more slowly than the old one, so programs may take longer to start.
1006  We believe the code quality is produces is about the same, so once
1007  started, programs should run at about the same speed.  Feedback about
1008  this would be useful.
1009
1010  On the plus side, Vex and hence Memcheck tracks value flow properly
1011  through floating point and vector registers, something the 2.X line
1012  could not do.  That means that Memcheck is much more likely to be
1013  usably accurate on vectorised code.
1014
1015- There is a subtle change to the way exiting of threaded programs
1016  is handled.  In 3.0, Valgrind's final diagnostic output (leak check,
1017  etc) is not printed until the last thread exits.  If the last thread
1018  to exit was not the original thread which started the program, any
1019  other process wait()-ing on this one to exit may conclude it has
1020  finished before the diagnostic output is printed.  This may not be
1021  what you expect.  2.X had a different scheme which avoided this
1022  problem, but caused deadlocks under obscure circumstances, so we
1023  are trying something different for 3.0.
1024
1025- Small changes in control log file naming which make it easier to
1026  use valgrind for debugging MPI-based programs.  The relevant
1027  new flags are --log-file-exactly= and --log-file-qualifier=.
1028
1029- As part of adding AMD64 support, DWARF2 CFI-based stack unwinding
1030  support was added.  In principle this means Valgrind can produce
1031  meaningful backtraces on x86 code compiled with -fomit-frame-pointer
1032  providing you also compile your code with -fasynchronous-unwind-tables.
1033
1034- The documentation build system has been completely redone.
1035  The documentation masters are now in XML format, and from that
1036  HTML, PostScript and PDF documentation is generated.  As a result
1037  the manual is now available in book form.  Note that the
1038  documentation in the source tarballs is pre-built, so you don't need
1039  any XML processing tools to build Valgrind from a tarball.
1040
1041Changes that are not user-visible:
1042
1043- The code has been massively overhauled in order to modularise it.
1044  As a result we hope it is easier to navigate and understand.
1045
1046- Lots of code has been rewritten.
1047
1048BUGS FIXED:
1049
1050110046  sz == 4 assertion failed
1051109810  vex amd64->IR: unhandled instruction bytes: 0xA3 0x4C 0x70 0xD7
1052109802  Add a plausible_stack_size command-line parameter ?
1053109783  unhandled ioctl TIOCMGET (running hw detection tool discover)
1054109780  unhandled ioctl BLKSSZGET (running fdisk -l /dev/hda)
1055109718  vex x86->IR: unhandled instruction: ffreep
1056109429  AMD64 unhandled syscall: 127 (sigpending)
1057109401  false positive uninit in strchr from ld-linux.so.2
1058109385  "stabs" parse failure
1059109378  amd64: unhandled instruction REP NOP
1060109376  amd64: unhandled instruction LOOP Jb
1061109363  AMD64 unhandled instruction bytes
1062109362  AMD64 unhandled syscall: 24 (sched_yield)
1063109358  fork() won't work with valgrind-3.0 SVN
1064109332  amd64 unhandled instruction: ADC Ev, Gv
1065109314  Bogus memcheck report on amd64
1066108883  Crash; vg_memory.c:905 (vgPlain_init_shadow_range):
1067        Assertion `vgPlain_defined_init_shadow_page()' failed.
1068108349  mincore syscall parameter checked incorrectly
1069108059  build infrastructure: small update
1070107524  epoll_ctl event parameter checked on EPOLL_CTL_DEL
1071107123  Vex dies with unhandled instructions: 0xD9 0x31 0xF 0xAE
1072106841  auxmap & openGL problems
1073106713  SDL_Init causes valgrind to exit
1074106352  setcontext and makecontext not handled correctly
1075106293  addresses beyond initial client stack allocation
1076        not checked in VALGRIND_DO_LEAK_CHECK
1077106283  PIE client programs are loaded at address 0
1078105831  Assertion `vgPlain_defined_init_shadow_page()' failed.
1079105039  long run-times probably due to memory manager
1080104797  valgrind needs to be aware of BLKGETSIZE64
1081103594  unhandled instruction: FICOM
1082103320  Valgrind 2.4.0 fails to compile with gcc 3.4.3 and -O0
1083103168  potentially memory leak in coregrind/ume.c
1084102039  bad permissions for mapped region at address 0xB7C73680
1085101881  weird assertion problem
1086101543  Support fadvise64 syscalls
108775247   x86_64/amd64 support (the biggest "bug" we have ever fixed)
1088
1089(3.0RC1: 27 July   05, vex r1303, valgrind r4283).
1090(3.0.0:   3 August 05, vex r1313, valgrind r4316).
1091
1092
1093
1094Stable release 2.4.1 (1 August 2005)
1095~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1096(The notes for this release have been lost.  Sorry!  It would have
1097contained various bug fixes but no new features.)
1098
1099
1100
1101Stable release 2.4.0 (March 2005) -- CHANGES RELATIVE TO 2.2.0
1102~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
11032.4.0 brings many significant changes and bug fixes.  The most
1104significant user-visible change is that we no longer supply our own
1105pthread implementation.  Instead, Valgrind is finally capable of
1106running the native thread library, either LinuxThreads or NPTL.
1107
1108This means our libpthread has gone, along with the bugs associated
1109with it.  Valgrind now supports the kernel's threading syscalls, and
1110lets you use your standard system libpthread.  As a result:
1111
1112* There are many fewer system dependencies and strange library-related
1113  bugs.  There is a small performance improvement, and a large
1114  stability improvement.
1115
1116* On the downside, Valgrind can no longer report misuses of the POSIX
1117  PThreads API.  It also means that Helgrind currently does not work.
1118  We hope to fix these problems in a future release.
1119
1120Note that running the native thread libraries does not mean Valgrind
1121is able to provide genuine concurrent execution on SMPs.  We still
1122impose the restriction that only one thread is running at any given
1123time.
1124
1125There are many other significant changes too:
1126
1127* Memcheck is (once again) the default tool.
1128
1129* The default stack backtrace is now 12 call frames, rather than 4.
1130
1131* Suppressions can have up to 25 call frame matches, rather than 4.
1132
1133* Memcheck and Addrcheck use less memory.  Under some circumstances,
1134  they no longer allocate shadow memory if there are large regions of
1135  memory with the same A/V states - such as an mmaped file.
1136
1137* The memory-leak detector in Memcheck and Addrcheck has been
1138  improved.  It now reports more types of memory leak, including
1139  leaked cycles.  When reporting leaked memory, it can distinguish
1140  between directly leaked memory (memory with no references), and
1141  indirectly leaked memory (memory only referred to by other leaked
1142  memory).
1143
1144* Memcheck's confusion over the effect of mprotect() has been fixed:
1145  previously mprotect could erroneously mark undefined data as
1146  defined.
1147
1148* Signal handling is much improved and should be very close to what
1149  you get when running natively.
1150
1151  One result of this is that Valgrind observes changes to sigcontexts
1152  passed to signal handlers.  Such modifications will take effect when
1153  the signal returns.  You will need to run with --single-step=yes to
1154  make this useful.
1155
1156* Valgrind is built in Position Independent Executable (PIE) format if
1157  your toolchain supports it.  This allows it to take advantage of all
1158  the available address space on systems with 4Gbyte user address
1159  spaces.
1160
1161* Valgrind can now run itself (requires PIE support).
1162
1163* Syscall arguments are now checked for validity.  Previously all
1164  memory used by syscalls was checked, but now the actual values
1165  passed are also checked.
1166
1167* Syscall wrappers are more robust against bad addresses being passed
1168  to syscalls: they will fail with EFAULT rather than killing Valgrind
1169  with SIGSEGV.
1170
1171* Because clone() is directly supported, some non-pthread uses of it
1172  will work.  Partial sharing (where some resources are shared, and
1173  some are not) is not supported.
1174
1175* open() and readlink() on /proc/self/exe are supported.
1176
1177BUGS FIXED:
1178
117988520   pipe+fork+dup2 kills the main program
118088604 	Valgrind Aborts when using $VALGRIND_OPTS and user progra...
118188614 	valgrind: vg_libpthread.c:2323 (read): Assertion `read_pt...
118288703 	Stabs parser fails to handle ";"
118388886 	ioctl wrappers for TIOCMBIS and TIOCMBIC
118489032 	valgrind pthread_cond_timedwait fails
118589106 	the 'impossible' happened
118689139 	Missing sched_setaffinity & sched_getaffinity
118789198 	valgrind lacks support for SIOCSPGRP and SIOCGPGRP
118889263 	Missing ioctl translations for scsi-generic and CD playing
118989440 	tests/deadlock.c line endings
119089481 	`impossible' happened: EXEC FAILED
119189663 	valgrind 2.2.0 crash on Redhat 7.2
119289792 	Report pthread_mutex_lock() deadlocks instead of returnin...
119390111 	statvfs64 gives invalid error/warning
119490128 	crash+memory fault with stabs generated by gnat for a run...
119590778 	VALGRIND_CHECK_DEFINED() not as documented in memcheck.h
119690834 	cachegrind crashes at end of program without reporting re...
119791028 	valgrind: vg_memory.c:229 (vgPlain_unmap_range): Assertio...
119891162 	valgrind crash while debugging drivel 1.2.1
119991199 	Unimplemented function
120091325 	Signal routing does not propagate the siginfo structure
120191599 	Assertion `cv == ((void *)0)'
120291604 	rw_lookup clears orig and sends the NULL value to rw_new
120391821 	Small problems building valgrind with $top_builddir ne $t...
120491844 	signal 11 (SIGSEGV) at get_tcb (libpthread.c:86) in corec...
120592264 	UNIMPLEMENTED FUNCTION: pthread_condattr_setpshared
120692331 	per-target flags necessitate AM_PROG_CC_C_O
120792420 	valgrind doesn't compile with linux 2.6.8.1/9
120892513 	Valgrind 2.2.0 generates some warning messages
120992528 	vg_symtab2.c:170 (addLoc): Assertion `loc->size > 0' failed.
121093096 	unhandled ioctl 0x4B3A and 0x5601
121193117 	Tool and core interface versions do not match
121293128 	Can't run valgrind --tool=memcheck because of unimplement...
121393174 	Valgrind can crash if passed bad args to certain syscalls
121493309 	Stack frame in new thread is badly aligned
121593328 	Wrong types used with sys_sigprocmask()
121693763 	/usr/include/asm/msr.h is missing
121793776 	valgrind: vg_memory.c:508 (vgPlain_find_map_space): Asser...
121893810 	fcntl() argument checking a bit too strict
121994378 	Assertion `tst->sigqueue_head != tst->sigqueue_tail' failed.
122094429 	valgrind 2.2.0 segfault with mmap64 in glibc 2.3.3
122194645 	Impossible happened: PINSRW mem
122294953 	valgrind: the `impossible' happened: SIGSEGV
122395667 	Valgrind does not work with any KDE app
122496243 	Assertion 'res==0' failed
122596252 	stage2 loader of valgrind fails to allocate memory
122696520 	All programs crashing at _dl_start (in /lib/ld-2.3.3.so) ...
122796660 	ioctl CDROMREADTOCENTRY causes bogus warnings
122896747 	After looping in a segfault handler, the impossible happens
122996923 	Zero sized arrays crash valgrind trace back with SIGFPE
123096948 	valgrind stops with assertion failure regarding mmap2
123196966 	valgrind fails when application opens more than 16 sockets
123297398 	valgrind: vg_libpthread.c:2667 Assertion failed
123397407 	valgrind: vg_mylibc.c:1226 (vgPlain_safe_fd): Assertion `...
123497427 	"Warning: invalid file descriptor -1 in syscall close()" ...
123597785 	missing backtrace
123697792 	build in obj dir fails - autoconf / makefile cleanup
123797880 	pthread_mutex_lock fails from shared library (special ker...
123897975 	program aborts without ang VG messages
123998129 	Failed when open and close file 230000 times using stdio
124098175 	Crashes when using valgrind-2.2.0 with a program using al...
124198288 	Massif broken
124298303 	UNIMPLEMENTED FUNCTION pthread_condattr_setpshared
124398630 	failed--compilation missing warnings.pm, fails to make he...
124498756 	Cannot valgrind signal-heavy kdrive X server
124598966 	valgrinding the JVM fails with a sanity check assertion
124699035 	Valgrind crashes while profiling
124799142 	loops with message "Signal 11 being dropped from thread 0...
124899195 	threaded apps crash on thread start (using QThread::start...
124999348 	Assertion `vgPlain_lseek(core_fd, 0, 1) == phdrs[i].p_off...
125099568 	False negative due to mishandling of mprotect
125199738 	valgrind memcheck crashes on program that uses sigitimer
125299923 	0-sized allocations are reported as leaks
125399949 	program seg faults after exit()
1254100036 	"newSuperblock's request for 1048576 bytes failed"
1255100116 	valgrind: (pthread_cond_init): Assertion `sizeof(* cond) ...
1256100486 	memcheck reports "valgrind: the `impossible' happened: V...
1257100833 	second call to "mremap" fails with EINVAL
1258101156 	(vgPlain_find_map_space): Assertion `(addr & ((1 << 12)-1...
1259101173 	Assertion `recDepth >= 0 && recDepth < 500' failed
1260101291 	creating threads in a forked process fails
1261101313 	valgrind causes different behavior when resizing a window...
1262101423 	segfault for c++ array of floats
1263101562 	valgrind massif dies on SIGINT even with signal handler r...
1264
1265
1266Stable release 2.2.0 (31 August 2004) -- CHANGES RELATIVE TO 2.0.0
1267~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
12682.2.0 brings nine months worth of improvements and bug fixes.  We
1269believe it to be a worthy successor to 2.0.0.  There are literally
1270hundreds of bug fixes and minor improvements.  There are also some
1271fairly major user-visible changes:
1272
1273* A complete overhaul of handling of system calls and signals, and
1274  their interaction with threads.  In general, the accuracy of the
1275  system call, thread and signal simulations is much improved:
1276
1277  - Blocking system calls behave exactly as they do when running
1278    natively (not on valgrind).  That is, if a syscall blocks only the
1279    calling thread when running natively, than it behaves the same on
1280    valgrind.  No more mysterious hangs because V doesn't know that some
1281    syscall or other, should block only the calling thread.
1282
1283  - Interrupted syscalls should now give more faithful results.
1284
1285  - Signal contexts in signal handlers are supported.
1286
1287* Improvements to NPTL support to the extent that V now works
1288  properly on NPTL-only setups.
1289
1290* Greater isolation between Valgrind and the program being run, so
1291  the program is less likely to inadvertently kill Valgrind by
1292  doing wild writes.
1293
1294* Massif: a new space profiling tool.  Try it!  It's cool, and it'll
1295  tell you in detail where and when your C/C++ code is allocating heap.
1296  Draws pretty .ps pictures of memory use against time.  A potentially
1297  powerful tool for making sense of your program's space use.
1298
1299* File descriptor leakage checks.  When enabled, Valgrind will print out
1300  a list of open file descriptors on exit.
1301
1302* Improved SSE2/SSE3 support.
1303
1304* Time-stamped output; use --time-stamp=yes
1305
1306
1307
1308Stable release 2.2.0 (31 August 2004) -- CHANGES RELATIVE TO 2.1.2
1309~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
13102.2.0 is not much different from 2.1.2, released seven weeks ago.
1311A number of bugs have been fixed, most notably #85658, which gave
1312problems for quite a few people.  There have been many internal
1313cleanups, but those are not user visible.
1314
1315The following bugs have been fixed since 2.1.2:
1316
131785658   Assert in coregrind/vg_libpthread.c:2326 (open64) !=
1318        (void*)0 failed
1319        This bug was reported multiple times, and so the following
1320        duplicates of it are also fixed: 87620, 85796, 85935, 86065,
1321        86919, 86988, 87917, 88156
1322
132380716   Semaphore mapping bug caused by unmap (sem_destroy)
1324        (Was fixed prior to 2.1.2)
1325
132686987   semctl and shmctl syscalls family is not handled properly
1327
132886696   valgrind 2.1.2 + RH AS2.1 + librt
1329
133086730   valgrind locks up at end of run with assertion failure
1331        in __pthread_unwind
1332
133386641   memcheck doesn't work with Mesa OpenGL/ATI on Suse 9.1
1334        (also fixes 74298, a duplicate of this)
1335
133685947   MMX/SSE unhandled instruction 'sfence'
1337
133884978   Wrong error "Conditional jump or move depends on
1339        uninitialised value" resulting from "sbbl %reg, %reg"
1340
134186254   ssort() fails when signed int return type from comparison is
1342        too small to handle result of unsigned int subtraction
1343
134487089   memalign( 4, xxx) makes valgrind assert
1345
134686407   Add support for low-level parallel port driver ioctls.
1347
134870587   Add timestamps to Valgrind output? (wishlist)
1349
135084937   vg_libpthread.c:2505 (se_remap): Assertion `res == 0'
1351        (fixed prior to 2.1.2)
1352
135386317   cannot load libSDL-1.2.so.0 using valgrind
1354
135586989   memcpy from mac_replace_strmem.c complains about
1356        uninitialized pointers passed when length to copy is zero
1357
135885811   gnu pascal symbol causes segmentation fault; ok in 2.0.0
1359
136079138   writing to sbrk()'d memory causes segfault
1361
136277369   sched deadlock while signal received during pthread_join
1363        and the joined thread exited
1364
136588115   In signal handler for SIGFPE,  siginfo->si_addr is wrong
1366        under Valgrind
1367
136878765   Massif crashes on app exit if FP exceptions are enabled
1369
1370Additionally there are the following changes, which are not
1371connected to any bug report numbers, AFAICS:
1372
1373* Fix scary bug causing mis-identification of SSE stores vs
1374  loads and so causing memcheck to sometimes give nonsense results
1375  on SSE code.
1376
1377* Add support for the POSIX message queue system calls.
1378
1379* Fix to allow 32-bit Valgrind to run on AMD64 boxes.  Note: this does
1380  NOT allow Valgrind to work with 64-bit executables - only with 32-bit
1381  executables on an AMD64 box.
1382
1383* At configure time, only check whether linux/mii.h can be processed
1384  so that we don't generate ugly warnings by trying to compile it.
1385
1386* Add support for POSIX clocks and timers.
1387
1388
1389
1390Developer (cvs head) release 2.1.2 (18 July 2004)
1391~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
13922.1.2 contains four months worth of bug fixes and refinements.
1393Although officially a developer release, we believe it to be stable
1394enough for widespread day-to-day use.  2.1.2 is pretty good, so try it
1395first, although there is a chance it won't work.  If so then try 2.0.0
1396and tell us what went wrong."  2.1.2 fixes a lot of problems present
1397in 2.0.0 and is generally a much better product.
1398
1399Relative to 2.1.1, a large number of minor problems with 2.1.1 have
1400been fixed, and so if you use 2.1.1 you should try 2.1.2.  Users of
1401the last stable release, 2.0.0, might also want to try this release.
1402
1403The following bugs, and probably many more, have been fixed.  These
1404are listed at http://bugs.kde.org.  Reporting a bug for valgrind in
1405the http://bugs.kde.org is much more likely to get you a fix than
1406mailing developers directly, so please continue to keep sending bugs
1407there.
1408
140976869   Crashes when running any tool under Fedora Core 2 test1
1410        This fixes the problem with returning from a signal handler
1411        when VDSOs are turned off in FC2.
1412
141369508   java 1.4.2 client fails with erroneous "stack size too small".
1414        This fix makes more of the pthread stack attribute related
1415        functions work properly.  Java still doesn't work though.
1416
141771906   malloc alignment should be 8, not 4
1418        All memory returned by malloc/new etc is now at least
1419        8-byte aligned.
1420
142181970   vg_alloc_ThreadState: no free slots available
1422        (closed because the workaround is simple: increase
1423         VG_N_THREADS, rebuild and try again.)
1424
142578514   Conditional jump or move depends on uninitialized value(s)
1426        (a slight mishanding of FP code in memcheck)
1427
142877952   pThread Support (crash) (due to initialisation-ordering probs)
1429        (also 85118)
1430
143180942   Addrcheck wasn't doing overlap checking as it should.
143278048   return NULL on malloc/new etc failure, instead of asserting
143373655   operator new() override in user .so files often doesn't get picked up
143483060   Valgrind does not handle native kernel AIO
143569872   Create proper coredumps after fatal signals
143682026   failure with new glibc versions: __libc_* functions are not exported
143770344   UNIMPLEMENTED FUNCTION: tcdrain
143881297   Cancellation of pthread_cond_wait does not require mutex
143982872   Using debug info from additional packages (wishlist)
144083025   Support for ioctls FIGETBSZ and FIBMAP
144183340   Support for ioctl HDIO_GET_IDENTITY
144279714   Support for the semtimedop system call.
144377022   Support for ioctls FBIOGET_VSCREENINFO and FBIOGET_FSCREENINFO
144482098   hp2ps ansification (wishlist)
144583573   Valgrind SIGSEGV on execve
144682999   show which cmdline option was erroneous (wishlist)
144783040   make valgrind VPATH and distcheck-clean (wishlist)
144883998   Assertion `newfd > vgPlain_max_fd' failed (see below)
144982722   Unchecked mmap in as_pad leads to mysterious failures later
145078958   memcheck seg faults while running Mozilla
145185416   Arguments with colon (e.g. --logsocket) ignored
1452
1453
1454Additionally there are the following changes, which are not
1455connected to any bug report numbers, AFAICS:
1456
1457* Rearranged address space layout relative to 2.1.1, so that
1458  Valgrind/tools will run out of memory later than currently in many
1459  circumstances.  This is good news esp. for Calltree.  It should
1460  be possible for client programs to allocate over 800MB of
1461  memory when using memcheck now.
1462
1463* Improved checking when laying out memory.  Should hopefully avoid
1464  the random segmentation faults that 2.1.1 sometimes caused.
1465
1466* Support for Fedora Core 2 and SuSE 9.1.  Improvements to NPTL
1467  support to the extent that V now works properly on NPTL-only setups.
1468
1469* Renamed the following options:
1470  --logfile-fd  -->  --log-fd
1471  --logfile     -->  --log-file
1472  --logsocket   -->  --log-socket
1473  to be consistent with each other and other options (esp. --input-fd).
1474
1475* Add support for SIOCGMIIPHY, SIOCGMIIREG and SIOCSMIIREG ioctls and
1476  improve the checking of other interface related ioctls.
1477
1478* Fix building with gcc-3.4.1.
1479
1480* Remove limit on number of semaphores supported.
1481
1482* Add support for syscalls: set_tid_address (258), acct (51).
1483
1484* Support instruction "repne movs" -- not official but seems to occur.
1485
1486* Implement an emulated soft limit for file descriptors in addition to
1487  the current reserved area, which effectively acts as a hard limit. The
1488  setrlimit system call now simply updates the emulated limits as best
1489  as possible - the hard limit is not allowed to move at all and just
1490  returns EPERM if you try and change it.  This should stop reductions
1491  in the soft limit causing assertions when valgrind tries to allocate
1492  descriptors from the reserved area.
1493  (This actually came from bug #83998).
1494
1495* Major overhaul of Cachegrind implementation.  First user-visible change
1496  is that cachegrind.out files are now typically 90% smaller than they
1497  used to be;  code annotation times are correspondingly much smaller.
1498  Second user-visible change is that hit/miss counts for code that is
1499  unloaded at run-time is no longer dumped into a single "discard" pile,
1500  but accurately preserved.
1501
1502* Client requests for telling valgrind about memory pools.
1503
1504
1505
1506Developer (cvs head) release 2.1.1 (12 March 2004)
1507~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
15082.1.1 contains some internal structural changes needed for V's
1509long-term future.  These don't affect end-users.  Most notable
1510user-visible changes are:
1511
1512* Greater isolation between Valgrind and the program being run, so
1513  the program is less likely to inadvertently kill Valgrind by
1514  doing wild writes.
1515
1516* Massif: a new space profiling tool.  Try it!  It's cool, and it'll
1517  tell you in detail where and when your C/C++ code is allocating heap.
1518  Draws pretty .ps pictures of memory use against time.  A potentially
1519  powerful tool for making sense of your program's space use.
1520
1521* Fixes for many bugs, including support for more SSE2/SSE3 instructions,
1522  various signal/syscall things, and various problems with debug
1523  info readers.
1524
1525* Support for glibc-2.3.3 based systems.
1526
1527We are now doing automatic overnight build-and-test runs on a variety
1528of distros.  As a result, we believe 2.1.1 builds and runs on:
1529Red Hat 7.2, 7.3, 8.0, 9, Fedora Core 1, SuSE 8.2, SuSE 9.
1530
1531
1532The following bugs, and probably many more, have been fixed.  These
1533are listed at http://bugs.kde.org.  Reporting a bug for valgrind in
1534the http://bugs.kde.org is much more likely to get you a fix than
1535mailing developers directly, so please continue to keep sending bugs
1536there.
1537
153869616   glibc 2.3.2 w/NPTL is massively different than what valgrind expects
153969856   I don't know how to instrument MMXish stuff (Helgrind)
154073892   valgrind segfaults starting with Objective-C debug info
1541        (fix for S-type stabs)
154273145   Valgrind complains too much about close(<reserved fd>)
154373902   Shadow memory allocation seems to fail on RedHat 8.0
154468633   VG_N_SEMAPHORES too low (V itself was leaking semaphores)
154575099   impossible to trace multiprocess programs
154676839   the `impossible' happened: disInstr: INT but not 0x80 !
154776762   vg_to_ucode.c:3748 (dis_push_segreg): Assertion `sz == 4' failed.
154876747   cannot include valgrind.h in c++ program
154976223   parsing B(3,10) gave NULL type => impossible happens
155075604   shmdt handling problem
155176416   Problems with gcc 3.4 snap 20040225
155275614   using -gstabs when building your programs the `impossible' happened
155375787   Patch for some CDROM ioctls CDORM_GET_MCN, CDROM_SEND_PACKET,
155475294   gcc 3.4 snapshot's libstdc++ have unsupported instructions.
1555        (REP RET)
155673326   vg_symtab2.c:272 (addScopeRange): Assertion `range->size > 0' failed.
155772596   not recognizing __libc_malloc
155869489   Would like to attach ddd to running program
155972781   Cachegrind crashes with kde programs
156073055   Illegal operand at DXTCV11CompressBlockSSE2 (more SSE opcodes)
156173026   Descriptor leak check reports port numbers wrongly
156271705   README_MISSING_SYSCALL_OR_IOCTL out of date
156372643   Improve support for SSE/SSE2 instructions
156472484   valgrind leaves it's own signal mask in place when execing
156572650   Signal Handling always seems to restart system calls
156672006   The mmap system call turns all errors in ENOMEM
156771781   gdb attach is pretty useless
156871180   unhandled instruction bytes: 0xF 0xAE 0x85 0xE8
156969886   writes to zero page cause valgrind to assert on exit
157071791   crash when valgrinding gimp 1.3 (stabs reader problem)
157169783   unhandled syscall: 218
157269782   unhandled instruction bytes: 0x66 0xF 0x2B 0x80
157370385   valgrind fails if the soft file descriptor limit is less
1574        than about 828
157569529   "rep; nop" should do a yield
157670827   programs with lots of shared libraries report "mmap failed"
1577        for some of them when reading symbols
157871028   glibc's strnlen is optimised enough to confuse valgrind
1579
1580
1581
1582
1583Unstable (cvs head) release 2.1.0 (15 December 2003)
1584~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1585For whatever it's worth, 2.1.0 actually seems pretty darn stable to me
1586(Julian).  It looks eminently usable, and given that it fixes some
1587significant bugs, may well be worth using on a day-to-day basis.
15882.1.0 is known to build and pass regression tests on: SuSE 9, SuSE
15898.2, RedHat 8.
1590
15912.1.0 most notably includes Jeremy Fitzhardinge's complete overhaul of
1592handling of system calls and signals, and their interaction with
1593threads.  In general, the accuracy of the system call, thread and
1594signal simulations is much improved.  Specifically:
1595
1596- Blocking system calls behave exactly as they do when running
1597  natively (not on valgrind).  That is, if a syscall blocks only the
1598  calling thread when running natively, than it behaves the same on
1599  valgrind.  No more mysterious hangs because V doesn't know that some
1600  syscall or other, should block only the calling thread.
1601
1602- Interrupted syscalls should now give more faithful results.
1603
1604- Finally, signal contexts in signal handlers are supported.  As a
1605  result, konqueror on SuSE 9 no longer segfaults when notified of
1606  file changes in directories it is watching.
1607
1608Other changes:
1609
1610- Robert Walsh's file descriptor leakage checks.  When enabled,
1611  Valgrind will print out a list of open file descriptors on
1612  exit.  Along with each file descriptor, Valgrind prints out a stack
1613  backtrace of where the file was opened and any details relating to the
1614  file descriptor such as the file name or socket details.
1615  To use, give: --track-fds=yes
1616
1617- Implemented a few more SSE/SSE2 instructions.
1618
1619- Less crud on the stack when you do 'where' inside a GDB attach.
1620
1621- Fixed the following bugs:
1622  68360: Valgrind does not compile against 2.6.0-testX kernels
1623  68525: CVS head doesn't compile on C90 compilers
1624  68566: pkgconfig support (wishlist)
1625  68588: Assertion `sz == 4' failed in vg_to_ucode.c (disInstr)
1626  69140: valgrind not able to explicitly specify a path to a binary.
1627  69432: helgrind asserts encountering a MutexErr when there are
1628         EraserErr suppressions
1629
1630- Increase the max size of the translation cache from 200k average bbs
1631  to 300k average bbs.  Programs on the size of OOo (680m17) are
1632  thrashing the cache at the smaller size, creating large numbers of
1633  retranslations and wasting significant time as a result.
1634
1635
1636
1637Stable release 2.0.0 (5 Nov 2003)
1638~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1639
16402.0.0 improves SSE/SSE2 support, fixes some minor bugs, and
1641improves support for SuSE 9 and the Red Hat "Severn" beta.
1642
1643- Further improvements to SSE/SSE2 support.  The entire test suite of
1644  the GNU Scientific Library (gsl-1.4) compiled with Intel Icc 7.1
1645  20030307Z '-g -O -xW' now works.  I think this gives pretty good
1646  coverage of SSE/SSE2 floating point instructions, or at least the
1647  subset emitted by Icc.
1648
1649- Also added support for the following instructions:
1650    MOVNTDQ UCOMISD UNPCKLPS UNPCKHPS SQRTSS
1651    PUSH/POP %{FS,GS}, and PUSH %CS (Nb: there is no POP %CS).
1652
1653- CFI support for GDB version 6.  Needed to enable newer GDBs
1654  to figure out where they are when using --gdb-attach=yes.
1655
1656- Fix this:
1657      mc_translate.c:1091 (memcheck_instrument): Assertion
1658      `u_in->size == 4 || u_in->size == 16' failed.
1659
1660- Return an error rather than panicing when given a bad socketcall.
1661
1662- Fix checking of syscall rt_sigtimedwait().
1663
1664- Implement __NR_clock_gettime (syscall 265).  Needed on Red Hat Severn.
1665
1666- Fixed bug in overlap check in strncpy() -- it was assuming the src was 'n'
1667  bytes long, when it could be shorter, which could cause false
1668  positives.
1669
1670- Support use of select() for very large numbers of file descriptors.
1671
1672- Don't fail silently if the executable is statically linked, or is
1673  setuid/setgid. Print an error message instead.
1674
1675- Support for old DWARF-1 format line number info.
1676
1677
1678
1679Snapshot 20031012 (12 October 2003)
1680~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1681
1682Three months worth of bug fixes, roughly.  Most significant single
1683change is improved SSE/SSE2 support, mostly thanks to Dirk Mueller.
1684
168520031012 builds on Red Hat Fedora ("Severn") but doesn't really work
1686(curiously, mozilla runs OK, but a modest "ls -l" bombs).  I hope to
1687get a working version out soon.  It may or may not work ok on the
1688forthcoming SuSE 9; I hear positive noises about it but haven't been
1689able to verify this myself (not until I get hold of a copy of 9).
1690
1691A detailed list of changes, in no particular order:
1692
1693- Describe --gen-suppressions in the FAQ.
1694
1695- Syscall __NR_waitpid supported.
1696
1697- Minor MMX bug fix.
1698
1699- -v prints program's argv[] at startup.
1700
1701- More glibc-2.3 suppressions.
1702
1703- Suppressions for stack underrun bug(s) in the c++ support library
1704  distributed with Intel Icc 7.0.
1705
1706- Fix problems reading /proc/self/maps.
1707
1708- Fix a couple of messages that should have been suppressed by -q,
1709  but weren't.
1710
1711- Make Addrcheck understand "Overlap" suppressions.
1712
1713- At startup, check if program is statically linked and bail out if so.
1714
1715- Cachegrind: Auto-detect Intel Pentium-M, also VIA Nehemiah
1716
1717- Memcheck/addrcheck: minor speed optimisations
1718
1719- Handle syscall __NR_brk more correctly than before.
1720
1721- Fixed incorrect allocate/free mismatch errors when using
1722  operator new(unsigned, std::nothrow_t const&)
1723  operator new[](unsigned, std::nothrow_t const&)
1724
1725- Support POSIX pthread spinlocks.
1726
1727- Fixups for clean compilation with gcc-3.3.1.
1728
1729- Implemented more opcodes:
1730    - push %es
1731    - push %ds
1732    - pop %es
1733    - pop %ds
1734    - movntq
1735    - sfence
1736    - pshufw
1737    - pavgb
1738    - ucomiss
1739    - enter
1740    - mov imm32, %esp
1741    - all "in" and "out" opcodes
1742    - inc/dec %esp
1743    - A whole bunch of SSE/SSE2 instructions
1744
1745- Memcheck: don't bomb on SSE/SSE2 code.
1746
1747
1748Snapshot 20030725 (25 July 2003)
1749~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1750
1751Fixes some minor problems in 20030716.
1752
1753- Fix bugs in overlap checking for strcpy/memcpy etc.
1754
1755- Do overlap checking with Addrcheck as well as Memcheck.
1756
1757- Fix this:
1758      Memcheck: the `impossible' happened:
1759      get_error_name: unexpected type
1760
1761- Install headers needed to compile new skins.
1762
1763- Remove leading spaces and colon in the LD_LIBRARY_PATH / LD_PRELOAD
1764  passed to non-traced children.
1765
1766- Fix file descriptor leak in valgrind-listener.
1767
1768- Fix longstanding bug in which the allocation point of a
1769  block resized by realloc was not correctly set.  This may
1770  have caused confusing error messages.
1771
1772
1773Snapshot 20030716 (16 July 2003)
1774~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1775
177620030716 is a snapshot of our current CVS head (development) branch.
1777This is the branch which will become valgrind-2.0.  It contains
1778significant enhancements over the 1.9.X branch.
1779
1780Despite this being a snapshot of the CVS head, it is believed to be
1781quite stable -- at least as stable as 1.9.6 or 1.0.4, if not more so
1782-- and therefore suitable for widespread use.  Please let us know asap
1783if it causes problems for you.
1784
1785Two reasons for releasing a snapshot now are:
1786
1787- It's been a while since 1.9.6, and this snapshot fixes
1788  various problems that 1.9.6 has with threaded programs
1789  on glibc-2.3.X based systems.
1790
1791- So as to make available improvements in the 2.0 line.
1792
1793Major changes in 20030716, as compared to 1.9.6:
1794
1795- More fixes to threading support on glibc-2.3.1 and 2.3.2-based
1796  systems (SuSE 8.2, Red Hat 9).  If you have had problems
1797  with inconsistent/illogical behaviour of errno, h_errno or the DNS
1798  resolver functions in threaded programs, 20030716 should improve
1799  matters.  This snapshot seems stable enough to run OpenOffice.org
1800  1.1rc on Red Hat 7.3, SuSE 8.2 and Red Hat 9, and that's a big
1801  threaded app if ever I saw one.
1802
1803- Automatic generation of suppression records; you no longer
1804  need to write them by hand.  Use --gen-suppressions=yes.
1805
1806- strcpy/memcpy/etc check their arguments for overlaps, when
1807  running with the Memcheck or Addrcheck skins.
1808
1809- malloc_usable_size() is now supported.
1810
1811- new client requests:
1812    - VALGRIND_COUNT_ERRORS, VALGRIND_COUNT_LEAKS:
1813      useful with regression testing
1814    - VALGRIND_NON_SIMD_CALL[0123]: for running arbitrary functions
1815      on real CPU (use with caution!)
1816
1817- The GDB attach mechanism is more flexible.  Allow the GDB to
1818  be run to be specified by --gdb-path=/path/to/gdb, and specify
1819  which file descriptor V will read its input from with
1820  --input-fd=<number>.
1821
1822- Cachegrind gives more accurate results (wasn't tracking instructions in
1823  malloc() and friends previously, is now).
1824
1825- Complete support for the MMX instruction set.
1826
1827- Partial support for the SSE and SSE2 instruction sets.  Work for this
1828  is ongoing.  About half the SSE/SSE2 instructions are done, so
1829  some SSE based programs may work.  Currently you need to specify
1830  --skin=addrcheck.  Basically not suitable for real use yet.
1831
1832- Significant speedups (10%-20%) for standard memory checking.
1833
1834- Fix assertion failure in pthread_once().
1835
1836- Fix this:
1837    valgrind: vg_intercept.c:598 (vgAllRoadsLeadToRome_select):
1838              Assertion `ms_end >= ms_now' failed.
1839
1840- Implement pthread_mutexattr_setpshared.
1841
1842- Understand Pentium 4 branch hints.  Also implemented a couple more
1843  obscure x86 instructions.
1844
1845- Lots of other minor bug fixes.
1846
1847- We have a decent regression test system, for the first time.
1848  This doesn't help you directly, but it does make it a lot easier
1849  for us to track the quality of the system, especially across
1850  multiple linux distributions.
1851
1852  You can run the regression tests with 'make regtest' after 'make
1853  install' completes.  On SuSE 8.2 and Red Hat 9 I get this:
1854
1855     == 84 tests, 0 stderr failures, 0 stdout failures ==
1856
1857  On Red Hat 8, I get this:
1858
1859     == 84 tests, 2 stderr failures, 1 stdout failure ==
1860     corecheck/tests/res_search               (stdout)
1861     memcheck/tests/sigaltstack               (stderr)
1862
1863  sigaltstack is probably harmless.  res_search doesn't work
1864  on R H 8 even running natively, so I'm not too worried.
1865
1866  On Red Hat 7.3, a glibc-2.2.5 system, I get these harmless failures:
1867
1868     == 84 tests, 2 stderr failures, 1 stdout failure ==
1869     corecheck/tests/pth_atfork1              (stdout)
1870     corecheck/tests/pth_atfork1              (stderr)
1871     memcheck/tests/sigaltstack               (stderr)
1872
1873  You need to run on a PII system, at least, since some tests
1874  contain P6-specific instructions, and the test machine needs
1875  access to the internet so that corecheck/tests/res_search
1876  (a test that the DNS resolver works) can function.
1877
1878As ever, thanks for the vast amount of feedback :) and bug reports :(
1879We may not answer all messages, but we do at least look at all of
1880them, and tend to fix the most frequently reported bugs.
1881
1882
1883
1884Version 1.9.6 (7 May 2003 or thereabouts)
1885~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1886
1887Major changes in 1.9.6:
1888
1889- Improved threading support for glibc >= 2.3.2 (SuSE 8.2,
1890  RedHat 9, to name but two ...)  It turned out that 1.9.5
1891  had problems with threading support on glibc >= 2.3.2,
1892  usually manifested by threaded programs deadlocking in system calls,
1893  or running unbelievably slowly.  Hopefully these are fixed now.  1.9.6
1894  is the first valgrind which gives reasonable support for
1895  glibc-2.3.2.  Also fixed a 2.3.2 problem with pthread_atfork().
1896
1897- Majorly expanded FAQ.txt.  We've added workarounds for all
1898  common problems for which a workaround is known.
1899
1900Minor changes in 1.9.6:
1901
1902- Fix identification of the main thread's stack.  Incorrect
1903  identification of it was causing some on-stack addresses to not get
1904  identified as such.  This only affected the usefulness of some error
1905  messages; the correctness of the checks made is unchanged.
1906
1907- Support for kernels >= 2.5.68.
1908
1909- Dummy implementations of __libc_current_sigrtmin,
1910  __libc_current_sigrtmax and __libc_allocate_rtsig, hopefully
1911  good enough to keep alive programs which previously died for lack of
1912  them.
1913
1914- Fix bug in the VALGRIND_DISCARD_TRANSLATIONS client request.
1915
1916- Fix bug in the DWARF2 debug line info loader, when instructions
1917  following each other have source lines far from each other
1918  (e.g. with inlined functions).
1919
1920- Debug info reading: read symbols from both "symtab" and "dynsym"
1921  sections, rather than merely from the one that comes last in the
1922  file.
1923
1924- New syscall support: prctl(), creat(), lookup_dcookie().
1925
1926- When checking calls to accept(), recvfrom(), getsocketopt(),
1927  don't complain if buffer values are NULL.
1928
1929- Try and avoid assertion failures in
1930  mash_LD_PRELOAD_and_LD_LIBRARY_PATH.
1931
1932- Minor bug fixes in cg_annotate.
1933
1934
1935
1936Version 1.9.5 (7 April 2003)
1937~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1938
1939It occurs to me that it would be helpful for valgrind users to record
1940in the source distribution the changes in each release.  So I now
1941attempt to mend my errant ways :-)  Changes in this and future releases
1942will be documented in the NEWS file in the source distribution.
1943
1944Major changes in 1.9.5:
1945
1946- (Critical bug fix): Fix a bug in the FPU simulation.  This was
1947  causing some floating point conditional tests not to work right.
1948  Several people reported this.  If you had floating point code which
1949  didn't work right on 1.9.1 to 1.9.4, it's worth trying 1.9.5.
1950
1951- Partial support for Red Hat 9.  RH9 uses the new Native Posix
1952  Threads Library (NPTL), instead of the older LinuxThreads.
1953  This potentially causes problems with V which will take some
1954  time to correct.  In the meantime we have partially worked around
1955  this, and so 1.9.5 works on RH9.  Threaded programs still work,
1956  but they may deadlock, because some system calls (accept, read,
1957  write, etc) which should be nonblocking, in fact do block.  This
1958  is a known bug which we are looking into.
1959
1960  If you can, your best bet (unfortunately) is to avoid using
1961  1.9.5 on a Red Hat 9 system, or on any NPTL-based distribution.
1962  If your glibc is 2.3.1 or earlier, you're almost certainly OK.
1963
1964Minor changes in 1.9.5:
1965
1966- Added some #errors to valgrind.h to ensure people don't include
1967  it accidentally in their sources.  This is a change from 1.0.X
1968  which was never properly documented.  The right thing to include
1969  is now memcheck.h.  Some people reported problems and strange
1970  behaviour when (incorrectly) including valgrind.h in code with
1971  1.9.1 -- 1.9.4.  This is no longer possible.
1972
1973- Add some __extension__ bits and pieces so that gcc configured
1974  for valgrind-checking compiles even with -Werror.  If you
1975  don't understand this, ignore it.  Of interest to gcc developers
1976  only.
1977
1978- Removed a pointless check which caused problems interworking
1979  with Clearcase.  V would complain about shared objects whose
1980  names did not end ".so", and refuse to run.  This is now fixed.
1981  In fact it was fixed in 1.9.4 but not documented.
1982
1983- Fixed a bug causing an assertion failure of "waiters == 1"
1984  somewhere in vg_scheduler.c, when running large threaded apps,
1985  notably MySQL.
1986
1987- Add support for the munlock system call (124).
1988
1989Some comments about future releases:
1990
19911.9.5 is, we hope, the most stable Valgrind so far.  It pretty much
1992supersedes the 1.0.X branch.  If you are a valgrind packager, please
1993consider making 1.9.5 available to your users.  You can regard the
19941.0.X branch as obsolete: 1.9.5 is stable and vastly superior.  There
1995are no plans at all for further releases of the 1.0.X branch.
1996
1997If you want a leading-edge valgrind, consider building the cvs head
1998(from SourceForge), or getting a snapshot of it.  Current cool stuff
1999going in includes MMX support (done); SSE/SSE2 support (in progress),
2000a significant (10-20%) performance improvement (done), and the usual
2001large collection of minor changes.  Hopefully we will be able to
2002improve our NPTL support, but no promises.
2003
2004