This change removes backtrace line (posix_fadvise64.c) that is not always in
output of the failing test (fadvise64).
It fixes memcheck/tests/mips64/fadvise64 on MIPS64 platforms with glibc 2.27
or newer.
Patch by Dimitrije Nikolic.
C++14 introduces sized delete operators and Valgrind support is added
by 6ef6f73. In addition, stderr filter which is used by Regtest should
be able to recognize this particular form in error report (just like
the other kinds of delete operators) in order to provide uniform output.
Fixes memcheck/tests/mismatches failure on non C++14 builds.
Patch by Aleksandar Rikalo.
Adding MIPS N32 ABI support.
BZ issue - #345763.
Contributed and maintained by mulitple people over the years:
Crestez Dan Leonard, Maran Pakkirisamy, Dimitrije Nikolic,
Aleksandar Rikalo, Tamara Vlahovic.
Patch by Paul Floyd, with comments slightly updated.
At the same time, cleanup of the trailing whitespaces in the whole file.
memcheck/tests/mismatches now shows the 'operator delete(void*, unsigned long)'
in the stacktrace, so updated the test expected output.
* Addition of a new configure option --enable-lto=yes or --enable-lto=no
Default value is --enable-lto=no, as the build is significantly slower,
so is not appropriate for valgrind development : this should be used
only on buildbots and/or by packagers.
* Some files containins asm functions have to be compiled without lto:
coregrind/m_libcsetjmp.c
coregrind/m_main.c
If these are compiled with lto, that gives undefined symbols at link time.
The files to compile without lto are
coregrind/m_libcsetjmp.c
coregrind/m_main.c
To compile these files with other options, a noinst target lib is defined.
The objects of this library are then added to the libcoregrind.
* memcheck/mc_main.c : move the handwritten asm helpers to mc_main_asm.c.
This avoids undefined symbols on some toolchains. Due to this,
the preprocessor symbols that activate the fast or asm memcheck helpers
are moved to mc_include.h
Platforms with handwritten helpers will also have the memcheck primary
map defined non static.
* In VEX, auxprogs/genoffsets.c also has to be compiled without lto,
as the asm produced by the compiler is post-processed to produce
pub/libvex_guest_offsets.h. lto not producing asm means the generation
fails if we used -flto to compile this file.
* all the various Makefile*am are modified to use LTO_CFLAGS for
(most) targets. LTO_CFLAGS is empty when --enable-lto=no,
otherwise is set to the flags needed for gcc.
If --enable-lto=no, LTO_AR and LTO_RANLIB are the standard AR and RANLIB,
otherwise they are the lto capable versions (gcc-ar and gcc-ranlib).
* This has been tested on:
debian 9.4/gcc 6.3.0/amd64+x86
rhel 7.4/gcc 6.4.0/amd64
ubuntu 17.10/gcc 7.2.0/amd64+x86
fedora26/gcc 7.3.1/s390x
No regressions on the above.
memcheck/Makefile.am contains 2 CFLAGS modifications lines that
are not working.
Remove these confusing lines.
(I have checked that the proper flags are still used for the 2 involved files)
Older gcc (4.8) default to GNU C90. Causing:
dlclose_leak.c:14:5: error: ‘for’ loop initial declarations are only
allowed in C99 mode
Fix by declaring int i before the loop.
As reported by Matthias Schwarzott <zzam@gentoo.org>. Testcase patch from him. The fix is
for check_CFSI_related_invariants() to avoid checking for overlaps against DebugInfos that are
in 'archived' status, since -- if a previously dlopened-and-then-dlclosed object is later
re-dlopened -- this may cause an overlap between the active and archived DebugInfos, which
is of no consequence. If the kernel maps the object to the same VMA the second time around
then there will *certainly* be an overlap.
(from bug 385408 comment 0):
Valgrind currently lacks support for the z/Architecture vector "support"
instructions introduced with z13. These are documented in the
z/Architecture Principles of Operation, Eleventh Edition (March, 2015),
chapter 21: "Vector Overview and Support Instructions".
Bug 387664 changes the default settings for accurate definedness checking
for {Add,Sub}{32,64} and {CmpEQ,CmpNE}{8,16,32,64}. This fix updates the
vbit tester (memcheck/tests/vbit-test) to test the accurate versions of
these, and thereby fixes a regression caused by
e847cb5429927317023d8410c3c56952aa47fb08 as committed for bug 387664.
Memcheck tries to accurately track definedness at the bit level, at least
for scalar integer operations. For many operations it is good enough to use
approximations which may overstate the undefinedness of the result of an
operation, provided that fully defined inputs still produce a fully defined
output. For example, the standard analysis for an integer add is
Add#(x#, y#) = Left(UifU(x#, y#))
which (as explained in the USENIX 05 paper
http://valgrind.org/docs/memcheck2005.pdf) means: for an add, worst-case
carry propagation is assumed. So all bits to the left of, and including,
the rightmost undefined bit in either operand, are assumed to be undefined.
As compilers have become increasingly aggressive, some of these
approximations are no longer good enough. For example, LLVM for some years
has used Add operations with partially undefined inputs, when it knows that
the carry propagation will not pollute important parts of the result.
Similarly, both GCC and LLVM will generate integer equality comparisons with
partially undefined inputs in situations where it knows the result of the
comparison will be defined. In both cases, Memcheck's default strategies
give rise to false uninitialised-value errors, and the problem is getting
worse as time goes by.
Memcheck already has expensive (non-default) instrumentation for integer
adds, subtracts, and equality comparisons. Currently these are only used if
you specify --expensive-definedness-checks=yes, and in some rare cases to do
with inlined string operations, as determined by analysing the block to be
instrumented, and by default on MacOS. The performance hit from them can be
quite high, up to 30% lossage.
This patch makes the following changes:
* During instrumentation, there is much finer control over which IROps get
expensive instrumentation. The following groups can now be selected
independently for expensive or cheap instrumentation:
Iop_Add32
Iop_Add64
Iop_Sub32
Iop_Sub64
Iop_CmpEQ32 and Iop_CmpNE32
Iop_CmpEQ64 and Iop_CmpNE64
This makes it possible to only enable, on a given platform, only the minimal
necessary set of expensive cases.
* The default set of expensive cases can be set on a per-platform basis.
This is set up in the first part of MC_(instrument).
* There is a new pre-instrumentation analysis pass. It identifies Iop_Add32
and Iop_Add64 uses for which the expensive handling will give the same
results as the cheap handling. This includes all adds that are used only
to create memory addresses. Given that the expensive handling of adds is,
well, expensive, and that most adds merely create memory addresses, this
more than halves the extra costs of expensive Add handling.
* The pre-existing "bogus literal" detection (0x80808080, etc) pass
has been rolled into the new pre-instrumentation analysis.
* The --expensive-definedness-checks= flag has been changed. Before, it
had two settings, "no" and "yes", with "no" being the default. Now, it
has three settings:
no -- always use the cheapest handling
auto -- use the minimum set of expensive handling needed to get
reasonable results on this platform, and perform
pre-instrumentation analysis so as to minimise the costs thereof
yes -- always use the most expensive handling
The default setting is now "auto". The user-visible effect of the new
default is that there should (hopefully) be a drop in false positive rates
but (unfortunately) also some drop in performance.
Memcheck reports an error on "if (n == 42)" in this test. Unless, that is,
accurate CmpEQ/NE definedness tracking is enabled. If you stare at this
long enough it is possible to see that the test "n == 42" isn't actually
undefined, because |n| is only ever zero or one, and only its least
significant bit is undefined. So the equality comparison against 42 is
defined because there are corresponding bits in the two operands that are
different and are both defined.
This commit fixes that by comparing with 1, which forces the result to
really depend on the only undefined bit in |n|.
I also added robustification:
* return arbitrary values from gcc_cant_inline_me(), so as to avoid gcc
simply copying the input to the output or otherwise deleting the
conditional branch.
* marking gcc_cant_inline_me() as un-inlineable
* Putting compiler barriers in the second conditional in main(), so gcc
can't simply ignore the result of the call to gcc_cant_inline_me() and
then delete the call entirely.
Rearrange big sections in mc_translate.c, so that the "main" instrumentation
function is at the end of the file rather than in the middle. The previous
layout never made much sense. The new layout is, roughly:
* stuff for baseline (level 2, non-origin tracking) instrumentation
* stuff for origin tracking (level 3) instrumentation
* the "final tidying" pass
* the main instrumentation function (and soon, a new pre-instrumentation
analysis pass)
New Iops are defined:
Iop_Scale2_32Fx4, Iop_Scale2_64Fx2,
Iop_Log2_32Fx4, Iop_Log2_64Fx2,
Iop_F32x4_2toQ16x8, Iop_F64x2_2toQ32x4,
Iop_PackOddLanes8x16, Iop_PackEvenLanes8x16,
Iop_PackOddLanes16x8, Iop_PackEvenLanes16x8,
Iop_PackOddLanes32x4, Iop_PackEvenLanes32x4.
Contributed by:
Tamara Vlahovic, Aleksandar Rikalo and Aleksandra Karadzic.
Related BZ issue - #382563.
The bug itself was solved in 3.12 by the addition of __gnu_cxx::__freeres
in the libstdc++ and have valgrind calling it before exit.
However, depending on the version of the libstdc++, the test leak_cpp_interior
was giving different results.
This commit adds some filtering specific to the test, so as to not depend
anymore of the absolute number of bytes leaked, and adds a suppression entry to
ignore the memory allocated by libstdc++.
This allows to have only 2 .exp files, instead of 4 (or worse, if
we would have to handle yet other .exp files depending on the libstdc++
version).
.. so that the code it creates runs in approximately half the time it did
before. This is in support of making the cost of expensive (exactly)
integer EQ/NE as low as possible, since the day will soon come when we'll
need to enable this by default.
Previous implementation misused some opcodes, and a side effect was
dead code emission.
To reimplement handling of these instructions, three new IoPs have been
introduced:
Iop_DivModU64to64, // :: I64,I64 -> I128
// of which lo half is div and hi half is mod
Iop_DivModS32to32, // :: I32,I32 -> I64
// of which lo half is div and hi half is mod
Iop_DivModU32to32, // :: I32,I32 -> I64
// of which lo half is div and hi half is mod
Patch by Aleksandra Karadzic and Tamara Vlahovic.
According to the epoll_pwait(2) man page:
The sigmask argument may be specified as NULL, in which case
epoll_pwait() is equivalent to epoll_wait().
But doing that under valgrind gives:
==13887== Syscall param epoll_pwait(sigmask) points to unaddressable byte(s)
==13887== at 0x4F2B940: epoll_pwait (epoll_pwait.c:43)
==13887== by 0x400ADE: main (syscalls-2007.c:89)
==13887== Address 0x0 is not stack'd, malloc'd or (recently) free'd
This is because the sys_epoll_pwait wrapper has:
if (ARG4)
PRE_MEM_READ( "epoll_pwait(sigmask)", ARG5, sizeof(vki_sigset_t) );
Which looks like a typo (ARG4 is timeout and ARG5 is sigmask).
This shows up with newer glibc which translates an epoll_wait call into
an epoll_pwait call with NULL sigmask.
Fix typo and add a testcase.
https://bugs.kde.org/show_bug.cgi?id=381289
git-svn-id: svn://svn.valgrind.org/valgrind/trunk@16451
Use .set noat directive to prevent assembler from complaining about
subsequent instructions that use register $at.
git-svn-id: svn://svn.valgrind.org/valgrind/trunk@16400
