Julian Seward [Sun, 21 Dec 2003 23:54:59 +0000 (23:54 +0000)]
Use the redir machinery to forcibly remap stpcpy in libc.so.6 to our
own version in mac_replace_strmem.c. We have to do this the hard way
because overenthusiastic PLT bypassing in glibc means the usual
symbol-override stuff doesn't work. IOW, for the usual reason that we
have to use the redir machinery at all.
This makes many programs run much more quietly on SuSE 9.
Make rep; nop (pause) yield the thread. Based on a patch by Tom Hughes;
I added a test case and cleaned up vg_dispatch.S while I was about it.
CCMAIL: 69529-done@bugs.kde.org
An experiment in generating branch-prediction hints. Enable them with
--branchpred=yes. I'm interested to know if these make a significant
difference for anyone - I see a small speed increase on the Pentium M.
This jumbo-checkin is the Full Virtualization checkin. This eliminates
Valgrind's dependency on the dynamic linker for getting started, and
instead takes things into its own hands.
This checkin doesn't add much in the way of new functionality, but it
is the basis for all future work on Valgrind. It allows us much more
flexibility in implementation, and well as increasing the reliability
of Valgrind by protecting it more from its clients.
This patch requires some changes to tools to update them to the changes
in the tool API, but they are straightforward. See the posting "Heads
up: Full Virtualization" on valgrind-developers for a more complete
description of this change and its effects on you.
When merging multiple symbols with the same address, rather than always
choosing the longest symbol, choose the longest ignoring any of the libc
junk prefixes like __libc_, __, __GI_*, etc. This makes the symbol
presented to the user in messages and used in *.supp files more consistent
and comprehensible.
Julian Seward [Sun, 14 Dec 2003 14:25:19 +0000 (14:25 +0000)]
Increase the max size of the translation cache from 200k average bbs to
300k average bbs. Programs on the size of OOo (680m17) are thrashing the
cache at the smaller size, creating large numbers of retranslations and
wasting significant time as a result.
Julian Seward [Sat, 13 Dec 2003 03:07:40 +0000 (03:07 +0000)]
Un-break 'make distcheck' yet again. I guess someone somewhere understands
how autoconf/automake really work, but not me :-) What's the difference
between $(srcdir) and $(top_srcdir) (eg in coregrind/Makefile.am) ?
Updated the README file. It had lots of out-of-date and incorrect information
in it, much of it from 1.0.X days. Did it in such a way that if it doesn't get
touched (and it undoubtedly won't) it won't really go out of date, eg. by
removing temporary details like version numbers, dates, details of specific
software incompatibilities. It's much better to be vague but correct, than
precise but incorrect; having incorrect info in a file as important as the
README is bad. Also removed the README_KDE3_FOLKS file because it's pretty
redundant now. Also added some changes that had been made in the stable branch
but not the HEAD.
Did similar, but smaller changes to README_DEVELOPERS and README_PACKAGERS.
Also updated the valgrind.spec.in file to use the new, post-1.0.X description
in the README.
Patch from Tom Hughes:
This patch extends the SFENCE support that is already present to include
support for LFENCE and MFENCE as well. It also stops CLFLUSH being mistaken
for SFENCE by checking the top two bits of the MODRM byte.
Tools using shadow memory can't handle the first 64KB being mapped, because
they rely on this area being unmapped for their quick sanity check. This
commit make Valgrind refuse to mmap() this area. Added a regtest for it.
Fixed up REP-prefix handling majorly. Factored out *lots* of repetitive code,
so much so that the file is now 280 lines shorter. This despite me also adding
support for LOOP{E,NE} (thanks to Abhijit Menon-Sen). Also added support for
CMPS[lw], which was missing. Adding more REP-prefix instructions in the future
will now be much easier.
As part of this, I moved the D-flag fetch outside of the REP loops. This might
make programs that use REP prefixes a lot go faster.
Dirk Mueller [Thu, 27 Nov 2003 02:16:41 +0000 (02:16 +0000)]
test for PARENT_SETTID support in clone() of the kernel instead
of testing for presence of NPTL by assuming that sys_futex is only
implemented when its a NPTL patched kernel.
Improved vg_libpthread's IGNORED/KLUDGED error messages; lots of people found
"kludged" confusing. Even printing out brief explanations of why functions are
ignored/kludged, in some cases.
Made the warning clearer when you try to catch SIGKILL/SIGSTOP. Also made it
clearer what's wrong if you try to catch signals 32 and 33; they're not bad
signals, just used internally. Updated one regtest accordingly.
Started to rescind the name "skin", replacing it with "tool". Did this in all
the places that normal users will see:
- command line: --tool=foo (although --skin=foo still works)
- docs: removed all traces (included renaming coregrind_skins.html to
coregrind_tools.html)
- in the usage messages
- in error messages
Also did in in some places that I judged were unlikely to cause clashes with
existing workspaces:
- in the header comments of many files (eg. "This file is part of Memcheck, a
Valgrind tool for...")
- in the regtests script
- in the .supp files
- in AUTHORS
- in README_MISSING_SYSCALL_OR_IOCTL
Greatly improved the symbol table reading code. It was a kludgy, accreted
mess. Pulled the symbol table reading bit out of vg_read_lib_symbols() into
read_symtab(), in the process removing the awful twice-only loop used to read
the symbol table and the dynamic symbol table. Factored out the code used to
find sections of interest (eg. .strtab, .symtab, .stabs, .plt, etc), replacing
nine (yes, nine) individual and slightly differing searches.
It's now much easier to understand, 132 lines shorter, and provides a better
base for easily doing more complicated debug stuff, eg. when we start reading
in all the debugging info (such as .debug_info for dwarf2).
Changed error messages: subsidiary explanations within an error message are now
indented one space; previously it was mixed, but Memcheck/Addrcheck indented
"Address" lines 3, which made them hard to see in the stack trace.
Changed error messages: now don't print anything below main() unless the new
option --show-below-main is on. It's on by default. It also affects
suppressions generated with --gen-suppressions=yes. Updated reg tests
accordingly.
Also updated docs for this. And added some missing command-line args to docs.
Also compartmentalised the options a little in the docs, and rearranged the
order of options in the usage message, in anticipation of a bigger
rearrangement that will be necessary soon -- to distinguish options used by all
skins from those used by error-checking skins, to skin-specific ones.
projects / thirdparty / valgrind.git / log
