Improve address description for address in the stack.
--read-var-info=yes is very memory and cpu intensive.
This patch ensures that even witout --read-var-info=yes that
the frame where the address point is reported in the address
description.
Julian Seward [Wed, 21 May 2014 14:43:11 +0000 (14:43 +0000)]
Add test cases for PCMPxSTRx cases 0x0E, 0x34, 0x14, and reformat some
of the associated switch statements. Pertains to #326469, #327639,
#328878 respectively.
on ppc64, pthread_create_WRK is not (always) produced in the stacktrace
showing where a thread was created.
This makes many tests fail => use sed to delete pthread_create_WRK
from the stacktrace to let tests succeed on ppc64.
With this change, on ppc64 gcc110 (fedora 18), helgrind failures
goes from 28 tests failing to 4, with following reasons:
helgrind/tests/pth_cond_destroy_busy (stderr)
(6 errors instead of 3 in the summary line ???)
helgrind/tests/tc06_two_races_xml (stderr)
similar change needed in filter_xml to del pthread_create_WRK
helgrind/tests/tc18_semabuse (stderr)
- with error code 22 (EINVAL: Invalid argument)
+ with error code 38 (ENOSYS: Function not implemented)
helgrind/tests/tc20_verifywrap (stderr)
- with error code 22 (EINVAL: Invalid argument)
+ with error code 38 (ENOSYS: Function not implemented)
More details about the stacktrace not containing pthread_create_WRK:
--------------------------------------------------------------------
Here is the stacktrace obtained by GDB+vgdb:
(gdb) bt
#0 0x0000008074f7ac5c in .__clone () from /lib64/libc.so.6
#1 0x000000807517b1ec in do_clone (pd=0x4c6f200, attr=0x8075189c90 <default_attr>, stackaddr=<optimized out>, stopped=<optimized out>,
fct=@0x80751a01e0: 0x807517c500 <start_thread>, clone_flags=4001536) at ../nptl/sysdeps/pthread/createthread.c:74
#2 0x000000000403ed0c in pthread_create_WRK (thread=<error reading variable: value has been optimized out>,
attr=<error reading variable: value has been optimized out>, start=<error reading variable: value has been optimized out>,
arg=0xfff00ee18) at hg_intercepts.c:269
#3 0x000000000403ef1c in _vgw00000ZZ_libpthreadZdsoZd0_pthreadZucreateZAZa (thread=<optimized out>, attr=<optimized out>,
start=<optimized out>, arg=<optimized out>) at hg_intercepts.c:300
#4 0x000000003806f1d8 in ?? ()
#5 0x0000008074e9fb94 in generic_start_main (main=@0x100200d8: 0x100013a0 <main>, argc=<optimized out>, ubp_av=0xfff00f2d8,
auxvec=0xfff00f408, init=<optimized out>, rtld_fini=<optimized out>, stack_end=<optimized out>, fini=<optimized out>)
at ../csu/libc-start.c:225
#6 0x0000008074e9fd90 in __libc_start_main (argc=<optimized out>, ubp_av=<optimized out>, ubp_ev=<optimized out>,
auxvec=<optimized out>, rtld_fini=<optimized out>, stinfo=<optimized out>, stack_on_entry=<optimized out>)
at ../sysdeps/unix/sysv/linux/powerpc/libc-start.c:91
#7 0x0000000000000000 in ?? ()
(gdb)
and here is the stacktrace produced by Valgrind unwinder:
Thread 1: status = VgTs_Runnable
==41687== at 0x8074F7AC5C: clone (in /usr/lib64/libc-2.16.so)
==41687== by 0x807517B1EB: do_clone.constprop.3 (createthread.c:74)
==41687== by 0x403EF1B: pthread_create@* (hg_intercepts.c:300)
==41687== by 0x10001597: main (tc19_shadowmem.c:172)
valgrind stack top usage: 15328 of 1048576
When the 2nd clone break is encountered (in the child thread), here is
the GDB stacktraces:
Thread 2 (Thread 6028):
#0 0x0000008074f75fb0 in .madvise () from /lib64/libc.so.6
#1 0x000000807517c700 in start_thread (arg=0x4c6f200) at pthread_create.c:402
#2 0x0000008074f7acf0 in .__clone () from /lib64/libc.so.6
Thread 1 (Thread 41687):
#0 pthread_create_WRK (thread=0xfff00ee10, attr=0x0, start=@0x100200e8: 0x10001dd0 <steer>, arg=0xfff00ee18) at hg_intercepts.c:248
#1 0x000000000403ef1c in _vgw00000ZZ_libpthreadZdsoZd0_pthreadZucreateZAZa (thread=<optimized out>, attr=<optimized out>,
start=<optimized out>, arg=<optimized out>) at hg_intercepts.c:300
#2 0x000000003806f1d8 in ?? ()
#3 0x0000008074e9fb94 in generic_start_main (main=@0x100200d8: 0x100013a0 <main>, argc=<optimized out>, ubp_av=0xfff00f2d8,
auxvec=0xfff00f408, init=<optimized out>, rtld_fini=<optimized out>, stack_end=<optimized out>, fini=<optimized out>)
at ../csu/libc-start.c:225
#4 0x0000008074e9fd90 in __libc_start_main (argc=<optimized out>, ubp_av=<optimized out>, ubp_ev=<optimized out>,
auxvec=<optimized out>, rtld_fini=<optimized out>, stinfo=<optimized out>, stack_on_entry=<optimized out>)
at ../sysdeps/unix/sysv/linux/powerpc/libc-start.c:91
#5 0x0000000000000000 in ?? ()
(gdb)
Here are the valgrind stacktraces:
Thread 1: status = VgTs_Runnable
==41687== at 0x403EBE0: pthread_create_WRK (hg_intercepts.c:248)
==41687== by 0x403EF1B: pthread_create@* (hg_intercepts.c:300)
==41687== by 0x8074E9FB93: generic_start_main.isra.0 (libc-start.c:225)
==41687== by 0x8074E9FD8F: (below main) (libc-start.c:91)
valgrind stack top usage: 15328 of 1048576
Thread 2: status = VgTs_WaitSys
==41687== at 0x8074F75FB0: madvise (in /usr/lib64/libc-2.16.so)
==41687== by 0x807517C6FF: start_thread (pthread_create.c:402)
valgrind stack top usage: 10320 of 1048576
And then after a few more next/breaks:
Thread 1: status = VgTs_Runnable
==41687== at 0x8074F7AC5C: clone (in /usr/lib64/libc-2.16.so)
==41687== by 0x807517B1EB: do_clone.constprop.3 (createthread.c:74)
==41687== by 0x403EF1B: pthread_create@* (hg_intercepts.c:300)
==41687== by 0x100015BB: main (tc19_shadowmem.c:173)
valgrind stack top usage: 15328 of 1048576
Thread 2: status = VgTs_WaitSys
==41687== at 0x8074F75FB0: madvise (in /usr/lib64/libc-2.16.so)
==41687== by 0x807517C6FF: start_thread (pthread_create.c:402)
valgrind stack top usage: 10320 of 1048576
So, pthread_create_WRK is not in the stacktrace anymore.
This only works in non-bi arch mode. If ever aarch64+arm
are compiled bi-arch, then some more work is needed to have
a 64 bits vgdb able to ptrace invoke a 32 bits valgrind.
Note also that PTRACE_GETREGSET is defined on other platforms
(e.g. ppc64 fedora 18 defines it), but it is not used on
these platforms, as again, PTRACE_GETREGSET implies some
work for bi-arch to work properly.
So, on all platforms except arm64, we use PTRACE_GETREGS
or PTRACE_PEEKUSER.
Bart Van Assche [Sat, 17 May 2014 10:44:00 +0000 (10:44 +0000)]
drd/tests/atomic_var: Revert r13876.
r13876 was a workaround for false ordering introduced by platform-specific
(Solaris) code. The conclusion of an off-list discussion was that this has
to be solved in the drd tool itself and not by modifying test programs. Hence
this revert.
On old kernel, poll syscall being ptraced (vgdb+ptrace) is not necessarily
properly restarted. Instead, it can fail with EINTR, even if no signal was
effectively received.
Handle such case by retrying the poll syscall when the poll syscall
is failing due to EINTR
Mark Wielaard [Fri, 16 May 2014 22:28:48 +0000 (22:28 +0000)]
Revert configure support for adding -Werror=format-security.
This reverts part of valgrind svn r13962. There was a typo in the configure
check that caused failures when -Werror=format-security wasn't supported
and the flag interfered badly with -Wno-format-zero-length. So remove
for now and only add back when properly tested on various (older) setups.
Julian Seward [Thu, 15 May 2014 13:50:47 +0000 (13:50 +0000)]
Make the PLAT_ identification work properly for mingw-win64. Problem was
that mingw64 also defines __MINGW32__, which led to the 32-bit definitions
being used in the 64-bit case. n-i-bz. (Bernhard.Loos@ruecker.de)
Julian Seward [Wed, 14 May 2014 23:38:23 +0000 (23:38 +0000)]
Implement VFPv4 VFMA and VFMS (F32 and F64 versions). Fixes #331057.
Patch from Janne Hellsten (jjhellst@gmail.com) with algebraic
rearrangement for the VFMS cases so as to make result signs match with
the hardware when some of the inputs are infinities.
Factorises the address code description and printing
of memcheck and helgrind in a common module:
pub_tool_addrinfo.h pub_core_addrinfo.h m_addrinfo.c
At the same time, the factorised code is made usable by other
tools also (and is used by the gdbserver command 'v.info location'
which replaces the helgrind 'describe addr' introduced 1 week ago
and which is now callable by all tools).
The new address description code can describe more addresses
(e.g. for memcheck, if the block is not on the free list anymore,
but is in an arena free list, this will also be described).
Similarly, helgrind address description can now describe more addresses
when --read-var-info=no is given (e.g. global symbols are
described, or addresses on the stack are described as
being on the stack, freed blocks in the arena free list are
described, ...).
See e.g. the change in helgrind/tests/annotate_rwlock.stderr.exp
or locked_vs_unlocked2.stderr.exp
The patch touches many files, but is basically a lot of improvements
in helgrind output files.
The code changes are mostly refactorisation of existing code.
Julian Seward [Tue, 13 May 2014 16:15:56 +0000 (16:15 +0000)]
Followup to r13958: add reg-trash lists to inline assembly in
TESTINSTPCMISALIGNED TESTINSTPCMISALIGNED_DWORDOUT
TESTINSTPCMISALIGNED_2OUT and nice up the the indentation a bit.
In case gdbsrv poll syscall fails, produces more information
gdbsrv poll syscall seems to very infrequently (1 on 100000 vgdb invocations§)
to be EINTR-upted.
So, when poll syscall fails, output stacktrace + sigmask status
to capture more info about the problem.
This is a follow-up/extension of r13748, which showed poll was interrupted
but it is not clear why. In particular, all async signals are supposed
to be masked at the time vgdb has forced an invocation
Julian Seward [Fri, 9 May 2014 16:13:21 +0000 (16:13 +0000)]
3_9_BUGSTATUS.txt: looked at all bugs in the file. Moved fixed ones
to NEWS (if not already there). Put the rest of them into a set of
categories depending on which part of the code base is affected, which
divides them up into -- IMO -- much more managable groups.
Mark Wielaard [Fri, 9 May 2014 13:34:13 +0000 (13:34 +0000)]
Out of tree build. Partial fix for Bug 333628.
Patch by Gilles Chanteperdrix <gilles.chanteperdrix@xenomai.org>.
Partial fix. make && make check now works with builddir != srcdir.
But make regtest doesn't yet.
Mark Wielaard [Fri, 9 May 2014 11:41:06 +0000 (11:41 +0000)]
Recognize MPX instructions and bnd prefix. Bug #333666.
Recognize and parse operands of new MPX instructions BNDMK, BNDCL,
BNDCU, BNDCN, BNDMOV, BNDLDX and BNDSTX. Also recognize bnd (F2) prefix
for CALL (E8,FF/2), RET (C2,C3), JMP (EB,E9,FF/4) and Jcc (70-7F,0F 80-8F).
All new MPX instructions are currently NOPs and the bnd prefix is ignored.
Keep the reading side of the FIFO opened after a vgdb disconnect
Otherwise, if the valgrind process has just forked before
the vgdb connect, and the child has not yet closed the FIFO it inheritates
from its parent, and a 2nd vgdb is started before the parent has
re-opened the reading side FIFO, the 2nd vgdb can connect to the child,
which will not handle the wakeup character/connection request, and
vgdb will have a write error and/or block.
projects / thirdparty / valgrind.git / log
