Julian Seward [Mon, 28 Feb 2011 09:03:44 +0000 (09:03 +0000)]
Don't construct the LAOG at all when --track-lockorders=no (as opposed
to previous behaviour, in which it was constructed but any resulting
errors were not shown, hence wasting CPU and memory.) Partial fix
for #255353. (Philippe Waroquiers, philippe.waroquiers@skynet.be)
Julian Seward [Sun, 27 Feb 2011 23:53:32 +0000 (23:53 +0000)]
Back out r11568 (Add a new constructor for empty XArrays,
VG_(newSizedXA)) since r11571 removes the only use of the
functionality that r11568 introduces.
Julian Seward [Sun, 27 Feb 2011 23:04:12 +0000 (23:04 +0000)]
Change the representation of VTSs. Instead of using an XArray of
ScalarTSs, have the ScalarTS array as a trailing array directly on the
VTS structure. This reduces the number of malloc'd blocks per VTS
from 3 to 1, since an XArray always requires 2 malloc'd blocks. At
least for tc19_shadowmem this reduces the total amount of heap
turnover in Arena 'tool' by a factor of 3, and modestly improves
performance whilst modestly reducing overall memory use.
Julian Seward [Thu, 24 Feb 2011 15:25:24 +0000 (15:25 +0000)]
Scalability fix for Helgrind: reduce the size of ScalarTS (scalar
timestamps) from 16 to 8 bytes. This halves the size of vector
timestamps and reduces the amount of memory needed to run programs
that have many threads and/or many synchronisation events.
The tradeoff is that Helgrind must abort the run if the program
creates more than 2^20 (1.0e+6) threads or performs more than 2^44
(1.76e+13) synchronisation events. Neither of these seem like a
significant limitation in practice. It's easy to argue that a limit
of 2^44 synch events would take at a minimum, several CPU months on a
very fast machine.
Julian Seward [Wed, 23 Feb 2011 13:30:53 +0000 (13:30 +0000)]
A scalability fix for Helgrind for running large workloads. When
creating new vector timestamps (VTSs) via tick and join operations,
preallocate the underlying XArray of ScalarTSs (scalar timestamps) at
the likely final size, using new function VG_(newSizedXA) introduced
in r11558. This reduces overall heap turnover (in VG_AR_TOOL) by a
factor of several. Together with revs 11567 and 11568, it mitigates
the worst-case performance falloff in long runs that involve lots of
threads and lots of synchronisation events (a.k.a Vector timestamps).
Julian Seward [Wed, 23 Feb 2011 13:22:24 +0000 (13:22 +0000)]
Add a new constructor for empty XArrays, VG_(newSizedXA). This is
identical to VG_(newXA) but allows passing in a size hint. In the
case where the likely final size of the XArray is known at creation
time, this allows avoiding the repeated (implicit) resizing and
copying of the array as elements are added, which can save a vast
amount of dynamic memory allocation turnover.
Julian Seward [Wed, 23 Feb 2011 13:18:56 +0000 (13:18 +0000)]
Fix a scalability problem observed whilst running Helgrind on a large
workload: when scanning a freelist of a given size for a big-enough
block (to allocate), don't scan all the way around the list. Instead
give up after 100 blocks and try the freelist above. The pathological
case (as observed) is that the freelist contains tens of thousands of
blocks, but all are too small for the current request, hence they are
all visited pointlessly. If the new heuristic is used, the freelist
start point is moved along by one block, so that future searches
eventually inspect the entire freelist, just very slowly.
Also, some improvements to stats gathering, and rename of some
existing stats fields in struct Arena.
Julian Seward [Fri, 11 Feb 2011 16:47:03 +0000 (16:47 +0000)]
Make ld.so:index redir mandatory for glibc-2.12 and later, on x86-linux.
Also, improve the failure message a bit, so as to tell people what package
they need to install, in at least some cases.
Bart Van Assche [Thu, 10 Feb 2011 21:03:47 +0000 (21:03 +0000)]
DRD: don't inline pthread intercepts because in combination with the current fragile implementation of the CALL_FN_* macros inlining intercepts can easily trigger stack alignment errors on Darwin.
Julian Seward [Wed, 9 Feb 2011 12:47:23 +0000 (12:47 +0000)]
_pre_mem_asciiz handlers in both tools: don't segfault if passed an
obviously invalid address. Fixes #255009. Investigation & initial
patch by Philippe Waroquiers (philippe.waroquiers@skynet.be)
When unwinding needs to be done because the stack pointer is reset
(e.g. by a longjmp), it makes no sense to interprete the control
flow change as call, but should be seen as a return.
This indirectly fixes bug 246152. Unwinding potentially changes the
exec state, which is unique for threads, but also for signal handlers.
E.g. this is true for a longjmp out of a signal handler. Exec state
changes modify members of struct CLG_(current_state), such as
CLG_(current_state).bbcc and CLG_(current_state).jmps_passed, which
are backed in CLG_(setup_bbcc)() by last_bbcc and passed, respectivly.
On a exec state change, these local vars go out of sync, and lead
to invalid data passed to CLG_(push_call_stack)() for handling a call,
which triggered data corruption, and the symptoms seen in bug 246152.
As in the given situation, there is no call anymore, there is no call
into CLG_(push_call_stack)(), and the corruption (or since last commit
the failed assertion) is not triggered any more.
Better failed assertion then silent data corruption
This is part 1 of the fix to bug 246152, and makes the bug
reproducable as failed assertion also on Ubuntu 10.10 on 64bit
machines. However, the test needs to be compiled 32bit (-m32).
Julian Seward [Thu, 27 Jan 2011 23:56:36 +0000 (23:56 +0000)]
Somewhat reduce the amount of mempool sanity checking, so as to avoid
rendering the mempool machinery impossibly slow for pools containing
many blocks. Fixes #255966.
If Massif's --threshold value was less than 1.0, in lines like this:
->00.00% (0B) in 11 places, all below massif's threshold (00.00%)
the threshold would always be incorrectly printed as 00.00%. This was
because the percentage printing was broken for percentages less than 1.0.
This change fixes this problem, and modifies a test to check for it.
Julian Seward [Sun, 23 Jan 2011 20:45:53 +0000 (20:45 +0000)]
Improve error reports for addressing errors in the presence of
mempools: try and relate an invalid address to known mempool
allocated blocks, and if that fails, to malloc'd blocks that
back the mempool. See #254420.
Print a stack trace as part of the "unhandled instruction bytes" warning.
Useful if the program in question catches signals, in which case the usual
"Process terminating..." stack trace isn't shown. Requested by Jesse
Ruderman.
Julian Seward [Mon, 10 Jan 2011 15:01:03 +0000 (15:01 +0000)]
Memcheck, None: update avg translation size to be more realistic.
Massif: specify avg translation size at all, so as to avoid excessive
retranslations caused by the fact that the default value is far below
reality for Massif.
When a shmat() size is passed to the tool, round it up to a page size. This
is how mmap() sizes are treated. It fixes an assertion failure in Massif
with --pages-as-heap=yes.
Julian Seward [Mon, 6 Dec 2010 11:40:04 +0000 (11:40 +0000)]
New command line option: --trace-children-skip-by-arg, which allows
chase/nochase decisions for child processes to be made on the basis
of their argv[] entries rather than on the name of their executables.
Julian Seward [Mon, 6 Dec 2010 11:11:29 +0000 (11:11 +0000)]
Minor improvements to PDB reading:
* better progress messages, to make it clear that reading of a
PDB is finished, and how much stuff was read from it
* don't mmap PDB files to read them -- instead use VG_(read).
This is because CIFS filesystem mounting only works reliably on
Linux when mounted with option '-o directio', and that
disallows mmap-ing files.
Julian Seward [Mon, 6 Dec 2010 10:56:09 +0000 (10:56 +0000)]
Speedups and fixes:
* (speedup) addMemEvent: generate inline code to check whether a
memory access is within 16k of the stack pointer, and if so
don't bother to call the helper
* (speedup) find_Block_containing: cache the most recently seen 2
blocks, and check new references in them first. This gives a
worthwhile speedup.
* (fix) at the end of the run, merge stats from un-freed blocks
back into APs. This fixes misleading stats that cause un-freed
blocks to appear to not have been accessed at all.
Tom Hughes [Fri, 12 Nov 2010 10:40:20 +0000 (10:40 +0000)]
Rework the strcasecmp stuff a little, based on Jakub Jelinek's patch
on bug #256600 because the original version turned out to somewhat
fragile across different glibc versions.
Bart Van Assche [Mon, 25 Oct 2010 18:18:54 +0000 (18:18 +0000)]
DRD: the help text does now indicate that lock contention detection is off by default / changed default value of the exclusive mutex threshold from 1000s to off. See also #255247.
projects / thirdparty / valgrind.git / log
