This patch further reduces the memory used by TT/TC (by about 15Mb
on 32 bits memcheck default nr of sectors).
Memory is reduced by using UShort typedef-s for Sector no and TTE no.
Note that for TTE no, we had a mixture of UShort, UInt and Int used
depending on the place (a TTE no was in any case constrained to be an UShort).
The bss memory/startup space is also reduced by allocating the htt on demand
(like tt and tc), using mmap the first time a sector is initialised.
Changes:
* pub_core_transtab.h :
* 2 typedef to identify a sector and a tt entry (these 2 types are UShort)
* add 2 #define 'invalid values' for these types
* change the interface to use these types rather than UInt
* m_transtab.c
* use wherever relevant these 2 new types rather than UInt or UShort
* replace the use of -1 by INV_SNO or INV_TTE
* remove now useless typecast from Int/UInt to UShort for tte
* schedule.c: use the new types
Florian Krohm [Mon, 23 Mar 2015 17:13:04 +0000 (17:13 +0000)]
Add VG_(am_is_bogus_client_stack_pointer)(Addr).
The function is used in VG_(client_syscall) to avoid extending the stack
when it is clear that the current value of the stack pointer does not
point into a segment that looks like a stack segment.
See the comments in the code there.
As a side effect of this we can now revert r15018 which increased
the stack size of the alternate stack in memcheck/tests/sigaltstack.c.
The reason is that the belief at the time: "alternate stack is too small"
was not correct. What instead happened was that VG_(client_syscall) called
VG_(extend_stack) without need (the syscall was tgkill) and the new stack
pointer happened to be in a file segment.
In other words: the current stack pointer was still within the alternate
stack, i.e. the alternate stack was (barely) large enough.
Tom Hughes [Sun, 22 Mar 2015 11:01:58 +0000 (11:01 +0000)]
Include the platform name in the unhandled system call message
We often get bug reports for an unhandled system call which don't
make it clear what platform is in use, which makes it impossible
to know which system call it is.
343902 --vgdb=yes doesn't break when --xml=yes is used
Changes ensures that gdbserver is called also when xml is yes.
When gdbserver is set to yes, we have to temporarily reset
xml output to no, as gdbserver output (e.g. print the last error)
has to be printed to gdb.
Florian Krohm [Sat, 21 Mar 2015 10:58:37 +0000 (10:58 +0000)]
Change the GCC demangler to not use VLA. The rationale is that these VLAs
are allocated on the stack and they can become quite large - in particular
when the client is a C++ application using the Boost library.
In combination with the demanglers recursive nature this can quickly lead
to exhaustion of valgrind's per-thread stack (which cannot be dynamically
grown). Additionally, due to the large VLAs (I've seen a 32k array) we
could run out of stack space without issuing a prior warning and instead
just segfault.
Therefore this patch allocates these arrays on the heap and frees them
later. Basically this is a respin of Joseph's r10385.
Change TT/TC hashing data structure (decreases memory by 50MB for memcheck 32bits)
This patch changes the way the transtab entries hash table is done.
Currently, the hash table is an open hash table considered full at 65%.
This means that in average, 1 entry on 3 is unused.
(all the hash table memory will be 'active' for big applications,
as the active entries are normally reasonably distributed over the hash table).
The size of a transtab entry is significant (about 150 Bytes).
To avoid having 35% of the entries unused, the translation table
is split in 2:
An hash table, that will contain an index pointing at the transtab entries.
With this technique, we are adding a small hash table,
but we spare 35% of the translation table.
Performance measurements have shown no degradation,
and some platforms have better performance. Not too clear why,
probably this helps platforms with small caches ?).
Florian Krohm [Mon, 16 Mar 2015 22:03:42 +0000 (22:03 +0000)]
Increase the size of the alternate stack. It was too small.
This was found by accident and there is no known way to detect
an overflow of an alternate stack in the general case.
New Option --avg-transtab-entry-size=<number> can be used to tune
the size of the translation table sectors, either to gain memory
or to avoid too many retranslations.
Fix the following errors detected by makefile check
memcheck/tests/Makefile.am:1: error: wrap8.stderr.exp-ppc64 is missing in EXTRA_DIST
memcheck/tests/Makefile.am:1: error: wrap8.stdout.exp-ppc64 is missing in EXTRA_DIST
memcheck/tests/Makefile.am:1: error: wrap8.stdout.exp2 is in EXTRA_DIST but doesn't exist
memcheck/tests/Makefile.am:1: error: wrap8.stderr.exp2 is in EXTRA_DIST but doesn't exist
Florian Krohm [Sat, 14 Mar 2015 10:15:23 +0000 (10:15 +0000)]
Organise the expected output files for the wrap8 testcase.
There is special behaviout on ppc64 only. Let the filenames
reflect that. At the same time update the ppc specific
output to what it is. The important thing here is that the
stack overflow is detected. Everything else is effectively a
don't care. Should line numbers and such differ in the future
that should be filtered out.
Florian Krohm [Sat, 14 Mar 2015 09:44:04 +0000 (09:44 +0000)]
Update the ppc64 specific results to what they are.
The difference of the expected results as compared to other
platforms is
- Location 0x........ is 2 bytes inside local var "budget"
- declared at varinfo6.c:3115, in frame #2 of thread 1
+ Address 0x........ is on thread 1's stack
+ in frame #2, created by BZ2_blockSort (varinfo6.c:3107)
Should the stderr output of this testcase in the future
match the generic output (varinfo6.stderr.exp) then this is
another incarnation of
https://bugs.kde.org/show_bug.cgi?id=345121
Florian Krohm [Fri, 13 Mar 2015 13:50:08 +0000 (13:50 +0000)]
Sort locks by their guestaddr to make the error output independent
of the dynamically allocated Lock addresses.
This restores helgrind/tests/locked_vs_unlocked2.stderr.exp
from r14931.
While regtesting the patch I've observed intermittent failures
of helgrind/tests/hg05_race2 like so:
--- ../../helgrind/tests/hg05_race2.stderr.exp (revision 15001)
+++ ../../helgrind/tests/hg05_race2.stderr.exp (working copy)
@@ -26,8 +26,7 @@
at 0x........: th (hg05_race2.c:17)
by 0x........: mythread_wrapper (hg_intercepts.c:...)
...
- Location 0x........ is 0 bytes inside foo.poot[5].plop[11],
- declared at hg05_race2.c:24, in frame #x of thread x
+ Address 0x........ is on thread #x's stack
@@ -42,8 +41,7 @@
at 0x........: th (hg05_race2.c:17)
by 0x........: mythread_wrapper (hg_intercepts.c:...)
...
- Location 0x........ is 0 bytes inside foo.poot[5].plop[11],
- declared at hg05_race2.c:24, in frame #x of thread x
+ Address 0x........ is on thread #x's stack
m_transtab.c statistic/tracing
* common up the identical debug and clo_stat traces
* add in the stats the nr of sectors recycled
* add the avg translation size in each sector recycled
and in the final statistics
(no functional change)
Implement command line option --valgrind-stacksize=<number>
This allows to decrease memory usage when using many threads,
if no big stacksize is needed by Valgrind.
If needed (e.g. for demangling big c++ symbols), the V stacksize
can be increased.
Florian Krohm [Thu, 12 Mar 2015 18:56:21 +0000 (18:56 +0000)]
Fix two bugs:
(1) In r14664 VG_(get_fnname_if_entry) was changed to always
return a function name, even if that function was *not* an
entry. That broke callgrind and was also confusing because
it contradicts what "get_fnname_if_entry" suggests.
(2) In r14189 a function call was removed because it was considered
redundant which it was not.
Both bugs were hunted down by Joseph Weidendorfer.
Florian Krohm [Tue, 10 Mar 2015 20:46:58 +0000 (20:46 +0000)]
Issue a warning if a function has more than 5 million bytes of
code. Previously functions exceeding that size were observed in the
field. Assert for 100x that amount.
Florian Krohm [Tue, 10 Mar 2015 16:13:59 +0000 (16:13 +0000)]
Add support for building with -fsanitize=undefined.
- add configure option --enable-ubsan
- add __ubsan helpers (by Julian)
This requires gcc 4.9.2 or later. Not all platforms are supported, though.
With this change and VEX r3099 regression tests pass on amd64
with a valgrind compiled with -fsanitize=undefined.
Florian Krohm [Sat, 7 Mar 2015 23:01:14 +0000 (23:01 +0000)]
New function VG_(am_mmap_client_heap) which swallows
VG_(am_set_segment_isCH_if_SkAnonC).
Rename VG_(am_set_segment_hasT_if_client_segment) to
VG_(am_set_segment_hasT) passing in an address (because that function
cannot possible take a pointer to a *const* segment). Also assert that
the segment containing the address is a client segment. Everything else
is a bug.
update NEWS to indicate that
335907 segfault when running wine's ddrawex/tests/surface.c under valgrind
is assumed to be fixed either by a previous change in 3.10
and/or by the commit for 343173
Rhys Kidd [Sat, 7 Mar 2015 14:57:39 +0000 (14:57 +0000)]
n-i-bz: Replace non-POSIX bzero with proper memset. At least for internal-only, Darwin functionality. Picked up by cppcheck. No regressions within test suite.
Rhys Kidd [Sat, 7 Mar 2015 08:36:20 +0000 (08:36 +0000)]
Fix unhandled syscall: unix:348 (__pthread_chdir) and unhandled syscall: unix:349 (__pthread_fchdir) on OS X
bz#344512
- Support these two undocumented syscalls.
- New regression test case added.
Rhys Kidd [Sat, 7 Mar 2015 05:22:12 +0000 (05:22 +0000)]
Fix stack traces missing penultimate frame
bz#344560
- Also fixes memcheck/tests/badpoll test on OS X
- Problem occurs because the guest stack seen in a system call pre or post
function happens to not have a correct topmost stack frame, as Darwin system
call stubs do not start with the usual function prolog.
- New regression test case added.
- Thanks to Greg Banks for research, patch and test case.
Julian Seward [Thu, 5 Mar 2015 00:52:07 +0000 (00:52 +0000)]
Minor changes in an attempt to improve performance and reduce
the amount of file-reading resulting from DiImage-cache misses.
CACHE_N_ENTRIES:
Increase the DiImage cache size from 256KB to 8MB to deal with
drastically worse locality when reading inline info. The 256KB
setting dates from befre inline-info-reading days.
is_in_CEnt: remove a conditional branch from the hot path (of |get|,
effectively)
This fixes a helgrind crash detected on android.
Android bionic pthread lib unmaps the stack for detached threads
before exiting.
Helgrind tries to unwind the stack to record a 'read' after
the stack unmap, just before the exit syscall.
The unwind then causes a SEGV.
The solution consists in tightening the calculation of
the stack limits, so as to stop unwinding when no valid stack
can be found.
Regression test reproduces the same problem by simulating the
bionic behaviour on linux, using asm similar to bionic lib.
Florian Krohm [Tue, 3 Mar 2015 14:56:17 +0000 (14:56 +0000)]
Produce a user message in case of stack overflow.
Change VG_(extend_stack) and VG_(am_extend_into_adjacent_reservation_client)
accordingly.
Remove some redundant checking.
Add testcase.
Rhys Kidd [Sun, 1 Mar 2015 13:22:21 +0000 (13:22 +0000)]
Fix missing libobjc suppressions on OS X 10.10
bz#344702
- Update darwin14.supp suppressions
- Includes malloc_zone_malloc and malloc related issues in system libraries
Rhys Kidd [Fri, 27 Feb 2015 11:34:07 +0000 (11:34 +0000)]
Fix memcheck/tests/err_disable4 test on OS X
bz#344621
- Unnamed semaphores are not supported on OS X, must use named semaphores.
- To use named semaphores sem_open() instead of sem_init() utilised.
- Test case updated accordingly across all platforms.
Florian Krohm [Thu, 26 Feb 2015 21:48:19 +0000 (21:48 +0000)]
Simplify do_brk
- remove redundant asserts
- let VG_(am_extend_into_adjacent_reservation_client) worry about
- whether delta is too large
- whether the segment abutting this one exists and is a reservation
segment
The function already checks these things. No need to do it again here.
- do_brk does not need to know that a reservation segment must not
shrink beyond a single page. That detail ought to be hidden in
the address space manager.
Also, turn a few conditions into asserts.
Mark Wielaard [Wed, 25 Feb 2015 14:00:14 +0000 (14:00 +0000)]
Bug #344318 socketcall should wrap recvmmsg and sendmmsg
Some architectures, e.g. s390, don't have dedicated recvmmsg and sendmmsg
system calls, but use the socketcall multiplexing system call with
SYS_RECVMMSG or SYS_SENDMMSG (just like the accept4 systemcall can also
be called through socketcall). Create separate helpers for recvmmsg and
sendmmsg helpers that can be used by either the direct syscall or the
socket call.
Florian Krohm [Wed, 25 Feb 2015 10:06:06 +0000 (10:06 +0000)]
Change VG_(am_extend_map_client) as follows:
- Tighten up on asserts
- Simplify; as the function grows memory into a free segment, there
cannot possibly be any translations to be discarded. Free segments
do not have translations. sane_NSegment will make sure.
- Change the prototype to take in the start address of the mapping and
return a pointer to the resized segment. Previously, the code
ok = VG_(am_extend_map_client)( &d, old_seg, needL );
if (!ok)
goto eNOMEM;
VG_TRACK( new_mem_mmap, needA, needL,
old_seg->hasR, old_seg->hasW, old_seg->hasX,
was examining old_seg->hasR etc even though VG_(am_extend_map_client)
stated that *old_seg was invalid after the function returned.
That wasn't exactly a problem, but clearly looked wrong.
Julian Seward [Tue, 24 Feb 2015 12:24:35 +0000 (12:24 +0000)]
Handle new IROps added in vex r3092:
Iop_RecipStep64Fx2, Iop_RSqrtStep64Fx2
Iop_RSqrtEst64Fx2, Iop_RecipEst64Fx2
Iop_RecpExpF64, Iop_RecpExpF32
--This line, and those below, will be ignored--
Fix configure for Linux kernel >= 4.0-rc1
In addition raise the minimal Linux version to 2.6 as there is almost
no test coverage for 2.4 and 2.6 was released in 2003.
Florian Krohm [Fri, 20 Feb 2015 14:00:23 +0000 (14:00 +0000)]
Pass in a mask of segment kinds to VG_(get_segment_starts)
and VG_(am_get_segment_starts) to indicate which segments
should be collected. That should solve the following problem:
in m_main.c we used to:
for (i = 0; i < n_seg_starts; i++) {
Word j, n;
NSegment const* seg
= VG_(am_find_nsegment)( seg_starts[i] );
vg_assert(seg);
if (seg->kind == SkFileC || seg->kind == SkAnonC) {
...
// ... dynamic memory allocation for valgrind
...
}
This caused the vassert(seg) to fire because the dynamic memory
allocation further down the loop changed segments such that a
valgrind segment which used to be non-SkFree suddenly became
SkFree and hence VG_(am_find_nsegment) returned NULL. Whoom.
With this revision we only collect the segments we're really
interested in. For the example above that is all client segments.
So if V allocates memory -- fine. That will not change the layout
of client segments.
Julian Seward [Fri, 20 Feb 2015 12:29:59 +0000 (12:29 +0000)]
canonicaliseSymtab: fix silly logic that could cause the function to
loop forever. In particular, it will do that when two symbols have
exactly the same address range but differ in their .isText attribute.
Fixes #342117.
Escape newlines in command arguments for "cmd:" header field in dumps
We could do unescaping in callgrind_annotate, but a escaped command
even seems better there.
Julian Seward [Wed, 18 Feb 2015 12:57:06 +0000 (12:57 +0000)]
Improve string table reading for MSVC2010 compiled code. Also seems
to help with MSVC2013 compiled code. Variant of a patch from
Mark Browning (mabrowningrr@gmail.com). Fixes #211529.
Julian Seward [Tue, 17 Feb 2015 13:46:26 +0000 (13:46 +0000)]
ML_(find_name_of_pdb_file): bit a bit more careful when grepping the
output from /usr/bin/strings, so as to not get confused by substrings
".pdb" and ".PDB" when they don't appear at the end of a line.
projects / thirdparty / valgrind.git / log
